Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Brand AI-Readiness Audit

    1

    Audit a website to understand why AI assistants may struggle to discover, crawl, understand, trust, or accurately describe a brand.

    $9.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Brand AI-Readiness Audit

    Brand AI-Readiness Audit

    Example session with this skill installed

    Audit https://tonyschocolonely.com using Brand AI-Readiness Audit.

    Run the complete workflow across all six specialist audits using the orchestrator. Examine crawler access, content rendering and readability, structured data, fact extractability, freshness and corroboration, and visitor engagement.

    Produce a prioritised Markdown report with:

    • An executive summary and overall assessment
    • Evidence-backed findings with affected URLs and observed values
    • Severity, confidence, and business impact for each finding
    • Specific recommended fixes and implementation examples where appropriate
    • The three highest-priority actions to take first
    • Checks that could not be completed and why
    • A machine-readable JSON report alongside the Markdown report

    Respect robots.txt and the skill's read-only crawling restrictions. Do not invent findings when evidence is unavailable. Clearly distinguish measured results from inferences.

    • Read your context and instructions
    • Compiled the brand ai-readiness audit

    AI readiness audit: tonyschocolonely.com

    Audited 2026-09-24T16:14:53Z · 30 pages crawled · 9 findings · 5 recommendations

    The one-page version

    Every line here is written out again further down with the evidence under it, and every figure here is one this report states somewhere else. Read this page and act on it; read the rest when you want the working.

    What this site is. Tony’s Chocolonely Global at tonyschocolonely.com. It describes itself as: "We are a private family-owned company that manufactures, imports, and distributes top tier products, including Tony's Chocolonely" This audit read it as an online shop (medium confidence, from: the site sells from its own pages (product pages), it also names places to visit, which a retailer with showrooms or stores does), and the advice below is tailored to that kind of site, so if that reading is wrong the tailoring is too.

    The single worst thing. F-002 — 4 pages publish JSON-LD values escaped as HTML. Serialise JSON-LD values as JSON rather than running them through the template's HTML-escape filter. (JSON-LD: structured data written into a page's code for machines to read)

    • What to do first. 3 fixes, in this order.
      1. 4 pages publish JSON-LD values escaped as HTML — developer, under an hour.
      2. 5 of the 30 crawled pages have no top-level heading — developer, under an hour.
      3. 11 of the 28 distinct pages this audit read share a meta description with another page — content owner, half a day to a day.

    How much of the site was read. 30 URLs were crawled on this site and 30 pages came back with readable text. No page count below is out of more than those 30 URLs: where a check looked at one kind of page it says which kind, and the only bigger number in this report is the site's own sitemap, which is named as the sitemap where it appears. (sitemap: a file listing the site's pages for crawlers)

    • What part of the audit did not run. Every sub-skill ran.

    The short version

    No blocking problems. 6 medium-severity items are worth fixing to improve how often and how accurately the brand gets quoted.

    6 of the 14 actions below need somebody who can edit the files the site is built from - its theme or template, not a page in the admin screen. That is your developer, your agency, or whoever built the site. The rest you can do yourself.

    SeverityReported with confidenceWorth checking
    Critical00
    High00
    Medium15
    Low21
    For information00

    6 of the 9 findings below rest on a weaker reading, so they are listed under "Worth checking first" rather than beside the findings the audit read directly. They are still counted in this table.

    Start here

    3 fixes, in the order to do them. The ranking puts the most severe band first, and inside a band puts the fix that changes most for the least work first.

    4 pages publish JSON-LD values escaped as HTML (F-002) — Serialise JSON-LD values as JSON rather than running them through the template's HTML-escape filter.
    Who does it: developer. Roughly: under an hour.
    2.

    5 of the 30 crawled pages have no top-level heading (F-008) — Give each of these pages an H1 that names its subject.
    Who does it: developer. Roughly: under an hour.
    3. 11 of the 28 distinct pages this audit read share a meta description with another page (F-007) — Give every page a unique title and a meta description that states the page's specific fact.
    Who does it: content owner. Roughly: half a day to a day.

    Medium findings

    Real improvements, worth scheduling.

    Each finding below carries two labels. F-001, F-002 and so on are positions in this report: they run in order and they move when the report changes, so the same F-number in a later audit is usually a different problem. The short name printed beside it is the problem's own name and does not move - quote that one when you compare two audits of this site, or when you tell us a finding is wrong.

    F-002 — 4 pages publish JSON-LD values escaped as HTML

    Priority: schedule · confidence high · found by structured-data-audit · jsonld-values-escaped-as-html

    • Why this class of problem matters: A page that looks complete to a human can be empty to a machine.

    What we found. Example: https://tonyschocolonely.com/collections/all declares description as What&#39;s everyone buzzing about? It&#39;s our mellow yellow classic flavour, milk chocol.... Inside <script type="application/ld+json"> there is no HTML parsing, so a consumer reads those characters literally rather than the punctuation they stand for. Seen on 4 of the 30 pages this crawl read.

    Why it matters. The escaped text is on 4 of the 25 pages graded, and it is page names and descriptions rather than the organisation's own name - so what a consumer gets wrong is how it quotes those pages, with the character reference in them. Escaped values parse as valid JSON, so no validator reports them.

    What to do. Serialise JSON-LD values as JSON rather than running them through the template's HTML-escape filter. Who does it: developer. Roughly: under an hour.

    • In whatever writes this block, print each value as JSON rather than passing it through the filter that escapes text for HTML.
    • This site's templates are Liquid, where the JSON filter is {{ value | json }} and the HTML-escape filter it has to replace is {{ value | escape }}.
    • The escape filter is right for text that lands in HTML and wrong inside a script element, where the JSON parser is the only reader and it does not decode character references.
    • Re-check one of the named pages in View Source: the value should carry the punctuation itself, not &#39; or &amp;.
    • This edits a template file rather than a setting, so it is a code change - made through the Shopify admin, with no deployment, but a change to the theme all the same. Whoever looks after the site's theme makes it: in-house that is whoever edits templates, and otherwise the agency, freelancer or contractor who built the site - the name on your web-development invoices is usually the right person to ask.
    • Affected pages: https://tonyschocolonely.com/collections/all · https://tonyschocolonely.com/collections/big-bars · https://tonyschocolonely.com/collections/small-bars · https://tonyschocolonely.com/products/dark-milk-pretzel-toffee-42

    Low findings

    Hygiene. Cheap to fix, small individual gain.

    F-008 — 5 of the 30 crawled pages have no top-level heading

    Priority: when convenient · confidence medium · found by fact-extractability-audit · heading-structure-unclear

    What we found. 5 of 30 crawled pages have no H1. Examples: https://tonyschocolonely.com/pages/bean-tracker, https://tonyschocolonely.com/pages/clmrs, https://tonyschocolonely.com/pages/living-income-model, https://tonyschocolonely.com/pages/mission-lock, https://tonyschocolonely.com/pages/tonys-sourcing-principles.

    Where we looked. The h1 elements, including any alt text they carry, on every crawled page whose own title does not say it is an error page. That is one source, so this is reported at reduced confidence.

    Why it matters. Headings are how a machine works out which part of a long page answers which question. Broken structure means the whole page is treated as one undifferentiated block.

    What to do. Give each of these pages an H1 that names its subject. Who does it: developer. Roughly: under an hour.

    • Add one H1 to each page listed, stating what the page is about, not a slogan.
    • Where the page already shows its name in large text, change that element to an H1. Styling can stay the same; only the tag changes.
    • Where a template draws the page (a listing, a contact page), fix it in the template once rather than page by page.

    F-007 — 11 of the 28 distinct pages this audit read share a meta description with another page

    Priority: when convenient · confidence high · found by structured-data-audit · title-and-description-hygiene

    • Why this class of problem matters: Assistants quote what is easy to reach, read and lift verbatim.

    What we found. 11 of 28 page(s) share a meta description with another page (2 distinct description(s) reused); the most repeated is on 8 of them and reads "Tony's Chocolonely: fighting exploitation in cocoa – poverty, child labor and deforestati…". Affected pages include: https://tonyschocolonely.com/, https://tonyschocolonely.com/blogs/news, https://tonyschocolonely.com/collections/all, https://tonyschocolonely.com/collections/big-bars, https://tonyschocolonely.com/collections/small-bars. Counted over 28 distinct pages: of the 30 this crawl read, 2 (https://tonyschocolonely.com/products/dark-almond-sea-salt-51-180g, https://tonyschocolonely.com/products/extra-dark-chocolate-70-180g) repeat another crawled page, by its text or by the canonical it declares, and are counted once with it.

    Where we looked. The <title> element and the meta description tag on every crawled page, and 2 other sources, all listed in report.json.

    Why it matters. The title and description are the shortest description of a page a consumer sees, and are often quoted directly. Duplicates make different pages look like the same page.

    What to do. Give every page a unique title and a meta description that states the page's specific fact. Who does it: content owner. Roughly: half a day to a day.

    • Write a unique <title> of 15-90 characters for each page, leading with the specific subject rather than the brand name (a character in a script that writes without spaces, such as Japanese, counts as 2.5 because it carries that much more).
    • Write a unique meta description of 50-165 characters that states the concrete fact on the page (the date, the place, the answer).
    • Duplicated titles usually come from a template that omits the page variable; fix the template rather than each page.
    • A title that reads as a web address is the same fault one step further on: the template fell back to the page's URL. Give it the page's own name.
    • Affected pages: 11 in total; the 5 listed are the first in alphabetical order, not a ranking, and each is named above.

    Worth checking first

    Weaker readings. The audit thinks each of these is probably true, but each rests on an inference rather than on something it read directly, so check it against the site before you spend money on it. Nothing has been removed from the report to get here: these are counted in the totals above and written out in full below, they are simply not mixed in with the findings the audit is sure of.

    F-003 — The site's markup declares 2 spellings of its own brand name

    Medium severity if confirmed · priority schedule · confidence high · found by structured-data-audit · markup-declares-more-than-one-brand-name

    What we found. This site's structured data states its brand as "Tony’s Chocolonely" on 11 crawled pages (https://tonyschocolonely.com/collections/all, https://tonyschocolonely.com/collections/big-bars); "Tony’s Chocolonely Global" on the site's own Organization block. Every string quoted here was read out of the markup named beside it. A consumer reading the markup has to decide whether these are one company or 2. Checked 11 of the 30 pages this crawl read.

    Where we looked. The brand property of every Product, Service and other sellable JSON-LD or Microdata node on every crawled page, read as a bare string or..., and 2 other sources, all listed in report.json.

    Why it matters. Structured data is read as the site's own assertion, so two spellings of one brand are read as two assertions. Each halves the evidence for the other, and a consumer cannot tell that both name one company.

    What to do. Give the brand one spelling and emit that same string from the product template and the Organization block. Who does it: developer. Roughly: under an hour.

    F-001 — The homepage never says what the brand is, though https://tonyschocolonely.com/pages/store-locator does

    Medium severity if confirmed · priority schedule · confidence medium · found by fact-extractability-audit · definition-not-on-the-homepage

    What we found. The one sentence on this site that says what "Tony’s Chocolonely Global" is sits on https://tonyschocolonely.com/pages/store-locator, which is neither the homepage nor the about page - "We are a private family-owned company that manufactures, imports, and distributes top tier products, including Tony's Chocolonely". Nothing in the opening of https://tonyschocolonely.com/, in the rest of its text, or in the description its markup declares says what the brand is, so an assistant that fetches the homepage and reads no further has nothing to repeat. Examined the whole of 1 of the 30 pages this crawl read - the homepage - plus the page carrying the sentence. The readings this rests on are finite lists of shapes and words, so a value written in a form or a language those lists do not cover would not have been seen: this is what this audit could not find, and not a statement that the site does not publish it.

    Where we looked. The opening headings and first 150 words of the homepage, then the whole of its text, read for a sentence saying what this is, and 1 other source, all listed in report.json.

    Why it matters. An assistant asked about a brand fetches the homepage first and often only. A definition it can only reach by opening a blog post or scrolling past the main content is a definition it will not quote, and the brand gets described in whatever words somebody else used instead.

    What to do. Copy that sentence, or one like it, into the first paragraph of the homepage. Who does it: content owner. Roughly: under an hour.

    • Put the sentence in the first paragraph of https://tonyschocolonely.com/ as plain HTML text: "We are a private family-owned company that manufactures, imports, and distributes top tier products, including Tony's Chocolonely"
    • Keep the wording the same on both pages. Two different sentences describing the same brand give an assistant two answers and it will pick one at random.
    • Put it above the main content - the grid, the listing, the article - and outside any slider or carousel, so it is in the HTML a fetcher receives rather than in markup a browser assembles.
    • Repeat the same sentence verbatim on the about page, in the Organization description property, on your LinkedIn page and in your press kit.
    <h1>Tony’s Chocolonely Global</h1>
    <p><strong>We are a private family-owned company that manufactures, imports, and distributes top tier products, including Tony's Chocolonely</strong></p>
    

    F-004 — 1 page of questions and answers has no FAQPage markup

    Medium severity if confirmed · priority schedule · confidence high · found by structured-data-audit · no-faq-schema

    What we found. Pages that answer questions but declare no FAQPage type: https://tonyschocolonely.com/pages/faqs. Checked 1 of the 1 faq page this crawl read.

    Where we looked. The FAQPage and QAPage JSON-LD types on each page, and 3 other sources, all listed in report.json.

    Why it matters. An FAQPage block hands over question-and-answer pairs already shaped like the thing an assistant is trying to produce, which makes them unusually easy to quote.

    What to do. Wrap the existing questions and answers in FAQPage JSON-LD. Who does it: developer. Roughly: under an hour.

    • Add FAQPage markup to https://tonyschocolonely.com/pages/faqs by hand: none of its answers could be read off the page in full, so no block is given here. Write one Question per question the page asks, and copy the answer the page prints beneath it, word for word, into acceptedAnswer.text.
    • The answer text in the markup must match the answer on that page word for word; a mismatch is treated as spam by several consumers. A placeholder in the snippet marks text this audit could not confirm word for word on the page; write the real answer there rather than publishing the placeholder.
    • Phrase the questions the way a person would ask an assistant, not as the internal section headings a page of questions is usually written from.
    • This edits a template file rather than a setting - written out in full under F-002 above.

    F-005 — Organization markup is present but 1 high-value property is unusable

    Medium severity if confirmed · priority schedule · confidence medium · found by structured-data-audit · organization-schema-missing-properties

    What we found. https://tonyschocolonely.com/ declares Organization and omits url. The same markup is on 2 pages of the 25 crawled. Checked 2 of the 2 home pages this crawl read.

    Where we looked. Every property on the Organization block, including nested objects such as address and contactPoint. That is one source, so this is reported at reduced confidence.

    • Why it matters. url is what ties this identity to this site rather than to whoever copied the block.

    What to do. Fill in the Organization properties a consumer cannot read off this block: url. Who does it: developer. Roughly: under an hour.

    • Add the missing properties to the Organization block you already have: url.
    • The snippet below is an addition, not a replacement. Your block also declares slogan, which the snippet does not model; keep it exactly as it is now.
    • url is this site's own home page address.
    • Re-validate after the change.
    • Affected pages: https://tonyschocolonely.com/ · https://tonyschocolonely.com/es
    <script type="application/ld+json">
    {
      "@context": "https://schema.org",
      "@type": "Organization",
      "name": "Tony’s Chocolonely Global",
      "url": "https://tonyschocolonely.com/",
      "logo": "https://tonyschocolonely.com/cdn/shop/files/1536_90_90_kar_melkv2-nw.jpg?v=1713869815&width=1536",
      "description": "Want to join us? With incredibly tasty chocolate we show the world chocolate can be made differently.",
      "sameAs": [
        "https://twitter.com/tonychocolonely/",
        "https://www.facebook.com/TonysChocolonely/",
        "https://www.instagram.com/tonyschocolonely_nl/",
        "https://www.linkedin.com/company/tony's-chocolonely",
        "https://www.youtube.com/@tonychocolonely"
      ]
    }
    </script>
    

    F-006 — Deep pages give visitors no visible sense of where they are

    Medium severity if confirmed · priority schedule · confidence high · found by engagement-audit and structured-data-audit · deep-pages-have-no-breadcrumb

    • Why this class of problem matters: A visitor arriving mid-journey from an answer carries context the site never sees, and leaves unless the page orients them and offers a next step.

    What we found. 7 of 14 crawled deep pages show no breadcrumb trail and no link up to the section they sit in. 6 of them do carry BreadcrumbList structured data, which a machine reads and a visitor cannot see. Examples: https://tonyschocolonely.com/pages/store-locator, https://tonyschocolonely.com/products/dark-almond-sea-salt-51-180g, https://tonyschocolonely.com/products/dark-almond-sea-salt-51-47g, https://tonyschocolonely.com/products/dark-chocolate-70-small-bar, https://tonyschocolonely.com/products/dark-milk-pretzel-toffee-42. Also observed by structured-data-audit: 8 of 14 crawled deep pages declare no BreadcrumbList. Examples: https://tonyschocolonely.com/blogs/news/cheers-to-a-stellar-year-of-making-impact-in-cocoa, https://tonyschocolonely.com/blogs/news/chocolate-scorecard-2025, https://tonyschocolonely.com/blogs/news/it-s-now-or-never-farmers-need-the-industry-s-commitment, https://tonyschocolonely.com/blogs/news/new-henkie-honka-new-chapter, https://tonyschocolonely.com/blogs/news/the-chocolate-scorecard-is-in.

    Where we looked. A breadcrumb element named in each page's own markup, and 1 other source, all listed in report.json.

    Why it matters. Someone arriving from an answer lands deep with no history. A breadcrumb is the cheapest way to show them what section they are in and give them one click upward instead of a click back.

    What to do. Add a visible breadcrumb trail to deep page templates. Who does it: developer. Roughly: under an hour.

    • Add a breadcrumb above the H1 on whichever templates sit below the top level of this site: Home > Section > This page.
    • Make every step except the current page a working link.
    • Add BreadcrumbList markup mirroring the visible breadcrumb trail.
    • Include the home page as position 1 and the current page as the last item.
    • This edits a template file rather than a setting - written out in full under F-002 above.
    • Affected pages: 14 in total; the 5 listed are the first in alphabetical order, not a ranking, and each is named above.

    F-009 — Only 5 off-site profiles are linked from the crawled pages or declared in sameAs

    Low severity if confirmed · priority schedule · confidence high · found by freshness-corroboration-audit · thin-off-site-profile-footprint

    • Why this class of problem matters: Trust comes from agreement across independent sources.

    What we found. Distinct off-site profiles linked from the crawled pages or listed in sameAs: Facebook, Instagram, LinkedIn, X, YouTube. In our within-category study, brands assistants name linked a mean of 7.2 and comparable brands they do not name linked 4.6 - the only measure that pointed the same way in all six categories tested. That study is ours: 29 crawlable sites, and whether an assistant "names" a brand was a judgement we made rather than a systematic query, so treat it as a strong association and not a proven cause. https://tonyschocolonely.com/llms.txt names 6 off-site addresses, but each is the platform this site is built on, a document to fetch, or a company's front door rather than an account, so the site's own agent file adds nothing here.

    Where we looked. Links on the crawled pages whose path names an account, and 2 other sources, all listed in report.json.

    Why it matters. A fact stated only on the brand's own site is one source's word for it. Each additional profile that repeats the same name and description is another independent source agreeing. This is the strongest signal we measured, and it is almost entirely within a brand's control.

    What to do. Claim more of the places that independently confirm the brand exists, and list every one of them in Organization sameAs. Who does it: marketing. Roughly: half a day to a day.

    • Claim the profiles that apply to your category. Breadth is what matters here, not prestige: the brands assistants name are on more platforms, not better ones.
    • Cover the obvious ones first: Instagram, TikTok, YouTube, Pinterest, Facebook, the review site your buyers check before ordering, and every marketplace you already sell on.
    • Claim the company record too - a business database entry and an employer profile - so an assistant asked who is behind the brand finds an answer that is not the brand's own site.
    • Use the identical name and the identical one-sentence description on every profile. Wording that matches exactly is what makes them corroborate rather than merely coexist.
    • List every profile URL in the Organization sameAs array on your site.
    • Link back to the site from each profile, so the connection is stated from both ends.

    Beyond the defects

    These are not problems with the site. They are things that would raise the chance of being found, quoted and trusted, and they are included only where the audit saw the condition that makes them relevant.

    Corroborate the brand across independent sources

    Claim the authoritative profiles, check for and then link or create a Wikidata item, and list them all in sameAs. This crawl found 5 off-site profiles linked from the pages it read. In our within-category study of 29 crawlable sites, 36 sampled, in six categories, brands assistants name link a mean of 7.2 and comparable brands they ignore link 4.6. That study is ours, so treat it as a strong association and not a proven cause.

    Why this works. Trust comes from agreement across independent sources — One site saying something is a claim. Several independent, verifiable sources saying the same thing is what makes it safe to repeat.

    • How to do it — marketing, roughly half a day to a day:
      • Claim or update the profiles that apply to you: the social platforms your buyers browse, the review site they check before ordering, and every marketplace you already sell on.
      • Search Wikidata for the name before creating anything: this audit could not run that lookup, so whether an item already exists is unknown here. If one exists, link it from the Organization sameAs array; if not, create it with the same one-sentence description you use on the site.
      • Claim the company record too - a business database entry and an employer profile - so an assistant asked who is behind this finds an answer that is not your own site.
      • List every profile URL in the Organization sameAs array.
      • Make each profile link back to the site, and name a real person on at least one of them with something checkable next to their name - a qualification, a registration number, how long they have done this. A set of accounts nobody is named on corroborates that the accounts exist, not that anybody is behind them.

    Write one canonical boilerplate and reuse it verbatim

    Agree a single paragraph describing what this is and use it, unchanged, everywhere. Start from the sentence the site already uses: "We are a private family-owned company that manufactures, imports, and distributes top tier products, including Tony's Chocolonely". This crawl found no Organization description carrying it.

    Why this works. Trust comes from agreement across independent sources — Identical wording across independent sources is the strongest agreement signal available, and it costs nothing but discipline.

    • How to do it — marketing, roughly under an hour:
      • Write one paragraph: what you are, who you serve, where you are, when you started.
      • Paste it, character for character, into the site, the Organization description, and Facebook, Instagram, LinkedIn, X and the others you have claimed.
      • Change it in one place and propagate; never let two versions coexist.
      • Review it once a year rather than rewriting it per channel.

    Write the comparison that concedes something

    Cover the framings people actually search - "X vs Y", "X for <use case>" - and be honest in them, because the honesty is what makes them usable as a source. https://tonyschocolonely.com/blogs/news/it-s-now-or-never-farmers-need-the-industry-s-commitment already compares: "That’s a 58% (!!) increase in farmers compared to the year before, as more brands joined Tony’s Open Chain.". That sentence concedes nothing, and none of the 30 pages this crawl read sets the comparison out.

    Why this works. Assistants personalise by context, so single-framing content narrows reach — Assistants answer the question as it was asked, and most buying questions are asked as a comparison. A brand with content only for its own framing is absent from all of them however good its own page is - and a page that only flatters itself is read as a claim rather than as evidence, so it is discounted even when it is found.

    • How to do it — content owner, roughly half a day to a day:
      • Say plainly who each option suits better, including where somebody else's is the right answer. A comparison that concedes nothing reads as marketing and is discounted; the version that concedes is the one that gets quoted, and it gets quoted on the part where you win.
      • Write one page per alternative people actually mention to you, not one page per competitor you can think of.
      • Put the concrete differences in a table - price, what it does, what it costs to start - so the comparison can be read without reading the prose.

    Give each page on this template its own facts

    2 pages here open with the same heading and the same first paragraph as another page of the same kind, so they are one template printed several times: https://tonyschocolonely.com/products/dark-almond-sea-salt-51-180g, https://tonyschocolonely.com/products/dark-almond-sea-salt-51-47g.

    Why this works. Assistants personalise by context, so single-framing content narrows reach — Answers are shaped by the asker's context, and a machine comparing two of these pages finds the same sentences on both. Pages that do not differ from each other are treated as one page, so only one of them is ever surfaced - and which one is not your choice.

    • How to do it — content owner, roughly half a day to a day:
      • Rewrite the H1 and the first paragraph of each of these pages to name the place, audience or use case it is for.
      • Put one fact in each that is true of it and false of the others: its own address and opening hours, its own price, the name of the person who runs it.
      • Leave the shared parts shared. The template is not the problem; the parts of it that should differ and do not are.

    Design newsletters and transactional email text-first

    Put the substance in the first two lines of readable text, never in an image. This crawl found an email sign-up form on 26 pages, such as https://tonyschocolonely.com/, https://tonyschocolonely.com/blogs/news, so this site sends email a machine will summarise.

    Why this works. AI email summaries are built from readable text — Inbox summaries are built from the text of a message. Substance carried in an image, or buried under filler, is simply absent from the summary the recipient reads.

    • How to do it — marketing, roughly under an hour:
      • Lead every email with two lines of plain text stating what it is and what changed.
      • Never put the main message in an image; use images to support text, not replace it.
      • Give every email a subject line that states the fact rather than teasing it.
      • Include a plain-text alternative part in every send.

    What an assistant would quote

    For each key page, the sentence an assistant fetching it would most likely lift. "Nothing quotable" means the page contains no self-contained sentence stating a fact.

    PageTypeLikely quote
    https://tonyschocolonely.com/homeNothing quotable — no sentence on this page both stands alone and states a fact, so an assistant fetching it would have nothing to quote
    https://tonyschocolonely.com/eshomeNothing quotable — no sentence on this page both stands alone and states a fact, so an assistant fetching it would have nothing to quote
    https://tonyschocolonely.com/pages/contactcontactNothing quotable — no sentence on this page both stands alone and states a fact, so an assistant fetching it would have nothing to quote
    https://tonyschocolonely.com/pages/faqsfaqWidespread deforestation.. and 1.56 million children* trapped in child labor.
    https://tonyschocolonely.com/pages/store-locatorlocationHanael food d.o.o. is the distributor and importer for many amazing brands including Tony's Chocolonely.
    https://tonyschocolonely.com/products/dark-almond-sea-salt-51-180gproductYou are currently on the dark almond sea salt 51% product page.
    https://tonyschocolonely.com/products/dark-almond-sea-salt-51-47gproductWe've used 34 cocoa beans for this product.
    https://tonyschocolonely.com/products/dark-chocolate-70-small-barproductYou are currently on the dark chocolate 70% product page.
    https://tonyschocolonely.com/products/dark-milk-pretzel-toffee-42productYou are currently on the dark milk pretzel toffee 42% product page.
    https://tonyschocolonely.com/products/extra-dark-chocolate-70-180gproductYou are currently on the extra dark chocolate 70% product page.
    https://tonyschocolonely.com/products/millk-32-caramel-sea-salt-180gproductYou are currently on the milk caramel sea salt 32% product page.

    Appendix: what was checked

    93 of the marketplace's 96 checks ran across 6 skills: 9 named by a finding above, 76 that did not apply, 7 that ran and found nothing wrong, and 0 that contributed to a finding without being named by one. A further 1 check fired, and no finding in this report is filed under it - the skill published none, or one of its findings was held back or merged away and which check belonged to it is not recorded - so it is not counted as contributing to one. The other 3 - robots-txt-rules, core-fact-founding-facts, core-fact-location-or-service-area - sit in branches of their skills that this site did not reach, so they were not asked here. The audit made 56 extra read-only requests beyond the crawl, used the user agent BrandAIReadinessAudit/1.0 (+read-only audit), and rendered in rendered mode.

    The page count was fixed by the plan rather than by where the clock fell, and the rounding is what lets two runs over a slightly different network agree. On that basis a second audit of this site, unchanged, reads the same pages and reports the same findings.

    • Note: 1 block(s) of text appear on at least 15 of the 30 readable pages and were treated as site furniture rather than as any one page's content

    Checks that did not apply, and why

    A check that stays quiet on purpose is as important as one that fires. Each of these ran, found nothing on this site for it to judge, and stopped there - which is not the same as looking and finding the site clean. Each reason is cut to its first clause; report.json keeps every word, under not_applicable:

    • crawl-access-audit
      • bot-manager-challenge-page: no page answered with a bot-manager verification page in place of its content

    bot-manager-user-agent-comparison: https://tonyschocolonely.com/ answered HTTP 200 the same way for this audit and for OAI-SearchBot (OpenAI

    canonical-host-consistency: every canonical URL uses a single hostname (canonical: the address a page names as its main version)

    • canonical-targets: all 30 declared canonical URLs are on this domain and resolve

    hreflang-between-language-editions: the crawled pages link 13 language hostnames (at.tonyschocolonely.com, be.tonyschocolonely.com, de.tonyschocolonely.com (hreflang: a tag telling machines which language edition of a page is which)

    • https-transport: the site is served over HTTPS

    llms-txt-presence: this site publishes https://tonyschocolonely.com/llms.txt, https://tonyschocolonely.com/llms-full.txt

    • meta-refresh-redirects: no crawled page redirects by meta refresh instead of an HTTP status
    • noindex-on-content-pages: no page this crawl read carries a noindex directive
    • non-200-rate: every fetched URL returned HTTP 200
    • origin-response-time: the median response across 30 page(s) was 0.7 s

    robots-agent-groups-unambiguous: each of the 7 user-agent groups in robots.txt names a token no other group names (robots.txt: the file at the site's root telling crawlers what they may fetch)

    robots-blocks-ai-answer-crawlers: nothing blocks the 13 search and live-fetch agents in this skill's crawler table

    robots-blocks-ai-training-crawlers: nothing blocks the 16 training crawlers and opt-out tokens in this skill's crawler table

    • robots-blocks-content-paths: one whole top-level section is closed to all crawlers, /policies/ - not reported

    robots-crawl-delay: robots.txt sets no Crawl-delay for the User-agent: * group, so a crawler is not asked to wait between requests

    robots-rules-are-path-shaped: every one of the 144 Allow/Disallow value(s) in this robots.txt is shaped like a URL path

    sitemap-excludes-private-paths: none of the 93 URLs listed in the sitemap is a cart, checkout, account, login, administration or wishlist address

    unknown-paths-return-200: a path that does not exist returned HTTP 404, so this site's missing pages are distinguishable from its real ones

    • engagement-audit
      • broken-internal-links: all 70 internal link targets tested resolved successfully

    consistent-site-chrome: all 25 crawled pages carry the header and footer links this site shares across its templates within their own language...

    dead-end-pages: every crawled content page offers a way onward: a link out of its main content, on this site or off it, a call to action

    form-friction: no enquiry form on this site requires more than 8 fields. 1 distinct enquiry form was measured, counted once each

    homepage-links-into-the-site: the homepage links 38 other addresses of this site, so a machine entering at the front page has somewhere to go from it

    homepage-orientation: the homepage leads with the H1 "Visit our sites" and offers "Visit Site" within the first 1500 characters

    • intrusive-interstitials: no page ships an open overlay in its initial markup
    • landing-page-answers: 6 of 6 crawled pages that publish a price also state at least one of a stock state

    landing-page-leads-with-the-price: 0 pages that publish a price in their own copy were crawled, fewer than the 3 it takes to say anything about a template

    • mobile-viewport: every crawled page declares a viewport meta tag in its delivered <head>

    onward-links-specific-to-the-page: 13 of 14 crawled deep pages carry at least one internal link that is not on nearly every other page of this site

    orphan-pages: the sitemap lists at least 93 URLs - the sitemap index names 9 sub-sitemaps and this audit read 5 of them

    page-headings-name-their-own-page: the 25 crawled content pages carrying an H1 carry 24 distinct headings between them

    page-weight: no page's delivered HTML exceeds 2,000,000 bytes, and no page loads scripts from more than 15 third-party hosts

    • primary-navigation: the primary navigation has 38 items with descriptive labels
    • readability: no long page runs without subheadings or contains repeated walls of text

    title-body-alignment: the title vocabulary of all 19 compared page(s) appears in their own body text or headings

    • too-many-top-level-headings: the most top-level headings on any crawled content page is 2 distinct headings
    • fact-extractability-audit

    answer-first-paragraphs: 4 of 4 sections on question-answering pages open with a sentence that asserts something rather than one that greets the...

    brand-naming-consistency: the site declares one name, in 4 authoritative places (jsonld:Organization, jsonld:Organization-off-home, og:site_name

    • core-facts-present: all 4 core facts are stated in plain text
    • entity-definition: a quotable definition exists on https://tonyschocolonely.com/pages/store-locator
    • headings-name-their-topic: no crawled page that is not an index of other pages has 5 or more H2 sections
    • long-sentences: fewer than 40% of sentences run over 30 words on every measurable page
    • freshness-corroboration-audit

    content-freshness: the newest article is dated 2026-05-08 (4 month(s) old) and 100% of dated articles are within 18 months

    date-signals-present: every page where a date is expected - articles, press pages, and pages in sections whose other pages are dated

    • entity-ambiguity: Wikidata's robots.txt closes its API to automated clients

    fact-consistency-across-pages: telephone, postal code and the Organization description are consistent wherever they appear

    • footer-copyright-year: no copyright year was found in the page text
    • profile-links-resolve: all 1 verifiable profile link(s) resolve (YouTube)
    • sitemap-lastmod-coverage: 97.8% of sitemap entries carry a <lastmod> value
    • sitemap-lastmod-recency: the newest sitemap <lastmod> is 2026-09-24 and the median is 2026-09-23
    • stale-year-references: no page claims to be current as of a year more than 2 years old
    • render-readability-audit

    client-rendered-chrome: each of the 25 content pages read here delivers its own navigation, footer or heading in the initial HTML response

    • crawlable-pagination: listing pages either paginate with real links or do not use a load-more control
    • facts-locked-in-images: no content page relies on imagery of its own to carry its main message

    facts-locked-in-pdfs: no page links to a pricing or specification PDF whose numbers are missing from the page text. no page links to a...

    • iframed-main-content: no page delegates its main content to an iframe

    image-alt-coverage: no image-heavy content page has images that carry neither an alt attribute nor any other description

    render-mode: this audit read the site with a browser, so the JavaScript gap below is measured rather than inferred

    • spa-shell-detection: every crawled content page delivered readable text in the initial HTML response

    static-vs-rendered-text-gap: on 5 rendered page(s) JavaScript added less than 60% of the text, so the delivered HTML already carries the content

    • thin-html: each of the 25 pages this check could assess carries at least 300 characters of body text
    • unrendered-template-artefact: no page prints a template expression, a merge field between sentinels
    • video-transcripts: no page relies on audio or video to carry content that is missing from its text
    • structured-data-audit
      • article-markup: all 7 article page(s) carry Article markup with dates and an author
      • canonical-declaration: no crawled page was reached at a second address differing only in its query parameters
      • homepage-title: the homepage is titled "Fighting exploitation in cocoa – Tony’s Chocolonely Global"
      • html-lang-attribute: every crawled page declares a lang attribute

    jsonld-matches-visible-text: the names and prices declared in JSON-LD match what the pages display, on the 2 crawled pages that declare one

    • listing-markup: no crawled page that the crawl typed as a product page turned out to be a list of other pages
    • microdata-and-rdfa: no page relies on microdata or RDFa in place of JSON-LD
    • offer-completeness: every sellable block on the 11 crawled pages that declare one carries a complete Offer
    • open-graph-tags: every crawled page carries og:title, og:description and og:image
    • per-location-markup: 0 page(s) carry an address of their own that is a place of this site's
    • product-markup: all 6 detected product page(s) carry Product markup
    • website-searchaction-markup: the homepage already declares WebSite markup

    Checks that ran and found nothing wrong

    Listed so that silence is never ambiguous. These ran against this site and were clean - they are not omissions:

    • homepage-reachable (crawl-access-audit)
    • redirect-chain-length (crawl-access-audit)
    • robots-blocks-all-crawlers (crawl-access-audit)
    • robots-txt-reachable (crawl-access-audit)
    • sitemap-parses (crawl-access-audit)
    • sitemap-present (crawl-access-audit)
    • sitemap-urls-resolve (crawl-access-audit)

    Checks that fired, with no finding above filed under them

    Each of these was marked as having found something by the skill that ran it, and no finding in this report is filed under its name. Either that skill published nothing at all, or something it raised did not survive into this report - held back as a question this audit could not answer, or merged into another finding - and which of its checks belonged to that one is not recorded, so none of them is claimed as contributing. They used to be listed under the heading above, which told the reader to go and look for a finding that is not there.

    • breadcrumb-markup (structured-data-audit)

    Observations merged

    • no-breadcrumb-schema from structured-data-audit was merged into deep-pages-have-no-breadcrumb, because both are one edit to one template.

    Generated by brand-ai-readiness-audit v1.0.0. Read-only: this audit made no change to the site and submitted no forms.

    every crawled page, and 2 other sources, all listed in report.json

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Brand AI-Readiness Audit combines seven coordinated skills into one evidence-based workflow. It crawls a website once, then runs six specialist audits over that same snapshot: crawler access, rendered content, structured data, fact extractability, freshness and corroboration, and visitor engagement.

    What you get

    • A prioritised report of findings and concrete recommendations, with the three fixes to do first.
    • Evidence and affected URLs behind every reported issue.
    • Who should do each fix (developer or content owner) and a rough time estimate.
    • Markdown and JSON reports, plus an optional shareable HTML report (--format html).
    • A read-only audit workflow: it inspects websites without modifying them.

    Requirements: Python 3.10+, the documented Python dependencies, and network access to the target website. Playwright is optional for rendered-DOM analysis.

    Best for: Developers, technical SEO practitioners, and teams investigating website discoverability and content accessibility for AI systems.

    The audit reports evidence-backed findings and its own limitations; it does not guarantee inclusion or ranking in AI-generated answers.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Marketing & SEO.

    • Results depend on which pages are reachable during the bounded crawl (at most 60 pages, about 5 minutes).
    • Sites that rely heavily on JavaScript may need the optional Playwright and Chromium install for accurate rendered-DOM analysis.
    • Robots.txt restrictions and access failures limit what can be concluded; unavailable evidence is reported as inconclusive, not as a defect.
    • Off-site corroboration checks are limited to sources the workflow can reach.
    • Findings are technical diagnostics and recommendations, not guarantees of visibility, citations, or ranking in AI-generated answers.
    • The audit does not modify the target website or apply fixes.
    • Sites behind a login cannot be audited.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 24 days ago

    Software
    Python 3.10+requestsbeautifulsoup4lxmlBrotliGit

    Permissions

    Terminal / Shell
    Read Files
    Network Access
    Browser

    File Scopes

    out/**
    skills/**

    Network access fetches the audited website read-only (GET and HEAD only, robots.txt obeyed, no logins or forms). Terminal runs the bundled Python scripts. Write access saves report.md, report.json and snapshot.json to the out/ folder. Nothing outside the skill folder and its out/ folder is written.

    Listed26 days ago
    Updated24 days ago

    What's inside

    Frequently Asked Questions

    It checks crawler access, content readability, structured data, fact extractability, freshness and corroboration, and visitor engagement, then combines the results into one prioritised report with evidence and recommended fixes.

    No. It identifies technical and content-related barriers and recommends improvements. AI systems' retrieval and ranking decisions are outside the tool's control.

    Python 3.10 or newer, the documented dependencies, a compatible agent able to execute the workflow, and network access to the target website. Playwright is optional for rendered-DOM analysis.

    No. It only sends GET and HEAD requests, obeys robots.txt, never logs in or submits forms, and writes its reports to your own machine.

    A prioritised Markdown report and a machine-readable JSON report. Add --format html for a shareable HTML version.

    Without Playwright it reads the HTML the server delivers. With the optional Playwright and Chromium install it also renders pages in a browser and compares the two.

    It checks whether a machine can get in, read the content, find labelled facts, quote a clear sentence about the brand, and trust that the information is current, then whether a human visitor stays. It reports what it could not check instead of treating missing evidence as a defect.

    New in Marketing & SEO