Your page tells one story. AI answers tell another.
A well-built page can still be missing from a recommendation. A familiar brand can appear even when its page has room to improve. That’s why we look at both.
Can AI read and understand your page?
We inspect the submitted page and supporting site files for structure, access, discovery and content signals. The result is a weighted diagnostic with findings you can act on.
Look inside the pageDoes AI recommend your brand in its answers?
We derive relevant questions from the business context, collect answers from selected AI engines and record ranked appearances, competing brands, citations and sentiment.
Look inside the answersReadiness measures page signals under our rubric. Visibility measures appearances in a small set of recorded answers. Improving one does not guarantee a change in the other.
From a URL to six lenses on your page.
We start with what’s there: the initial HTML, the rendered page, its text and markup, and the files that describe how the site can be discovered.
Fetch the page
Initial HTML and HTTP signals
Render and extract
Text, structure and metadata
Evaluate the evidence
Rule checks and AI assessments
Six categories. Selected checks. Evidence you can inspect.
Structured Data
Can machines identify what this page is about?
Four valid JSON-LD scripts. Organization and WebSite markup complete.
See checks and interpretation for Structured Data
What we check
JSON-LD, page metadata, entity links, date markup and media semantics.
How to interpret it
We inspect the markup and properties expected for the detected page type. This is a check of the page’s implementation, not a requirement imposed by every AI engine.
LLM-Friendly Formatting
Is the information easy to navigate and extract?
One H1, 15 H2s, 11 H3s and 42 HTML lists.
See checks and interpretation for LLM-Friendly Formatting
What we check
Heading structure, HTML and Markdown lists, tables and question-and-answer blocks.
How to interpret it
We check for structural signals that organize the page. A passing result does not mean that every heading or answer has been judged editorially perfect.
Accessibility
How much can a crawler read, and how does the page perform?
100% initial-to-rendered text ratio. Recorded LCP: 1.49 seconds.
See checks and interpretation for Accessibility
What we check
Text in initial versus rendered HTML, plus LCP, CLS and TTFB.
How to interpret it
The rendering check compares extracted text lengths, not a semantic match of every sentence. Performance uses desktop CrUX field data when available, with a desktop PageSpeed lab fallback. It is not a complete accessibility or Core Web Vitals certification.
LLM as a Judge
Does the content align with relevant questions and read clearly?
13 of 20 generated questions matched. Clarity: 9 out of 10.
See checks and interpretation for LLM as a Judge
What we check
Semantic alignment with generated questions and an LLM clarity assessment.
How to interpret it
The question-alignment check compares embeddings of questions and page text. Its threshold is calibrated per run and includes randomness, so this is a directional diagnostic. It is separate from asking ChatGPT or Gemini whether they recommend the brand.
Discoverability
What discovery and access signals can we observe?
Robots rules allowed the checked agents. MCP.json could not be fetched.
See checks and interpretation for Discoverability
What we check
Robots rules, sitemaps, HTTP hygiene, social metadata, llms.txt and MCP.json.
How to interpret it
Search crawlers, training crawlers and user-requested retrieval serve different purposes. Optional llms.txt and MCP.json checks describe our rubric; they do not establish a requirement for appearing in AI answers.
Readability
How demanding and focused is the page’s writing?
Recorded readability: 47; SMOG index: 13. Main-content ratio: 25%.
See checks and interpretation for Readability
What we check
Language-aware readability measures and main-content versus boilerplate density.
How to interpret it
Metrics are interpreted using the detected content language. They help identify writing to review, rather than prove that a model will misunderstand the page. Different languages use different supported measures.
Miniatures illustrate selected checks from the recorded sample. The category scores summarize the full set of included checks.
Open any lens to see what it checks, what Zapier’s sample recorded and how to interpret the result. Some checks can be disabled by configuration, so compare the included checks when comparing reports.
Zapier’s readiness assessment generated 20 questions and matched 13 using semantic similarity; it also recorded clarity at 9/10. These are separate from the five questions in the visibility audit. The similarity threshold is calibrated per run, including a randomized target, so “13 of 20 matched” is a directional signal, rather than proof that the page correctly answers 65% of customer questions.
How we calculate the AI readiness score
Readiness uses two levels of weighting. Individual checks build a category score. The available category scores then build the overall score.
A weight determines how much a category influences the overall score. Higher weights have more influence. The total is the sum of the included category weights—44 in this sample.
Σ(check score × check weight) ÷ Σ(10 × check weight) × 100Positive-weight checks use a 0–10 scale; the result is rounded to a category score out of 100. Skipped checks are removed before this calculation.
| Category | Score | Weight | Score × Weight |
|---|---|---|---|
| Structured DataChecks for structured data like Schema tags, page title, and publish dates - so AI systems understand your content and can cite it accordingly. | 96 | 10 | 960 |
| LLM-Friendly FormattingDetects FAQs, lists, and tables that help AI systems extract information and produce accurate answers. | 100 | 2 | 200 |
| AccessibilityConfirms that key content is available before JavaScript runs and remains accessible on mobile devices and to AI crawlers. | 100 | 10 | 1000 |
| LLM as a JudgeUses real-world prompts and an AI model to test how clearly your page answers user questions - measuring how well it speaks "AI". | 80 | 10 | 800 |
| DiscoverabilityChecks whether search and AI crawlers can discover your site through robots.txt, sitemaps, llms.txt, and valid status codes. | 87 | 6 | 522 |
| ReadabilityMeasures whether your content is clear, concise, and easy for people and AI systems to understand. | 50 | 6 | 300 |
| Weighted total | 44 | 3,782 |
(96 × 10 + 100 × 2 + 100 × 10 + 80 × 10 + 87 × 6 + 50 × 6) ÷ 44≈ 86The result contributes.
A measured failure can score zero. Partial credit and the assigned check weight affect its category.
Leave it out.
Skipped checks are excluded. A category with no included results is unavailable and omitted from the overall average.
Read the reason.
A positive-weight error recorded with score zero still contributes. Some unavailable checks return “skipped” instead. The status matters.
If a category has no included check results—for example, because its checks are disabled in the report configuration or all are skipped—its weight is excluded from the overall calculation. A category containing results but no positive weights receives zero. The 44-unit denominator above applies to this sample, where all six category scores are available.
Zapier’s 86 highlights a useful contrast: strong markup and formatting alongside a readability score of 50. The findings make that contrast actionable. The number alone cannot predict whether an AI engine will recommend Zapier.
How we measure AI search visibility
The page gives us the brand, business type and market context. We use that context to build relevant search questions, then review how AI platforms such as ChatGPT and Gemini answer them.
- 01Understand the business
Identify the brand, industry and relevant language or location from the submitted page.
- 02Generate the questions
Build a small set of category and customer-intent prompts. Review the actual wording in the report.
- 03Collect and interpret answers
Query the selected AI platforms, retain the returned evidence and extract the brand’s position and competing recommendations.
Which workflow automation platform is best for connecting AI tools with business apps?
ChatGPT
RECORDED ANSWER“For most businesses, Zapier is the best default”
Gemini
RECORDED ANSWER“Zapier is the undisputed king of SaaS integrations, boasting connections to over 9,000+ apps.”
Both count as an appearance. Their positions tell different stories.
Ranked appearances ÷ valid answer checks × 100For this sample: 9 ÷ 10 × 100 = 90. #1 and #8 both count as an appearance. Their positions remain visible, so the same score can describe very different competitive situations.
A valid check with no ranked appearance counts as an absence. Failed collection or unavailable ranking interpretation is excluded from the denominator; if there are no valid checks, the visibility calculation fails rather than reporting a meaningful zero.
For content sites and marketplaces, the current parser can use a citation-based position when a recommendation list is unavailable. Interpret the position alongside the business type and source evidence. Visibility is always a snapshot of the selected questions and engines.
Found in an answer. Recommended. Cited. Each tells you something different.
Follow the thread from a brand name to a recommendation to a source link. These observations can overlap, but they answer different questions.
ChatGPT
RECORDED EXCERPThelp.zapier.com“For most businesses, Zapier is the best default”
Mention
Zapier is named
Recommendation
Ranked #1
Citation
A source is linked
In the enterprise governance question, Gemini’s recorded answer cites Zapier’s site but does not give Zapier a ranked appearance. Workato is the first result. That answer contributes to own-site citation coverage, while remaining an absence in the visibility score.
Explore this in the sample reportBeing present is only part of the story.
Answer-level sentiment assesses the language used about the brand in each sampled response. Zapier’s sample has 4 positive and 5 mixed answers; the remaining answer has no ranked appearance.
Brand perception adds another perspective.
The report also researches strengths and concerns across web sources. Zapier’s sample records four of each. This assessment is separate from the ten answer-level labels and does not change the visibility score.
Keep the context next to the score.
An audit is a useful observation, with a scope and a moment in time. Knowing how it was collected helps you decide what to do with it.
A page, not every page.
Readiness checks your submitted page and supporting files. Visibility samples relevant questions. Neither covers your entire site or every AI search.
A market, where it matters.
Collection can target relevant countries and languages. Global software queries are usually location-neutral. Translating a report does not search a new market.
A snapshot, not a promise.
Questions, answers and scoring can change. Compare evidence and collection settings before linking a score change to a website edit.
A daily cache can be reused.
Matching questions and collection settings may reuse answers from the same day’s 24-hour cache. Rerunning does not always collect fresh answers.
Our discovery checks include llms.txt and MCP.json. Their presence does not establish a visibility benefit. Google states that it ignores llms.txt for Search visibility and requires no special schema for its generative AI features. Different crawler roles also matter: search indexing, training and user-requested retrieval are distinct.
Read Google’s AI optimization guidanceThe details behind the headline numbers.
Does a readiness score of 86 mean an 86% chance of appearing in AI search?
No. It is a weighted diagnostic score under AI Page Ready’s rubric. It does not estimate a search ranking, probability of recommendation or citation rate. The AI Search Visibility report separately records what happened in the sampled answers.
Does the audit cover the whole website?
The readiness audit evaluates the submitted page, with supporting checks of site-level files such as robots.txt and the sitemap. Visibility derives the brand and market context from that page and asks a small set of relevant questions. Neither is a comprehensive crawl of every page or a census of every possible AI answer.
What does #1 or #8 mean in the visibility report?
It is the position assigned to the brand in the extracted recommendation list for that particular recorded answer. It is not a permanent ChatGPT or Gemini ranking. Being named in prose, appearing in a recommendation list and having a URL cited are different observations.
Can an unchanged page receive a different score on another run?
Yes. Generated questions, AI judgments, changing answers, collection failures and per-run semantic threshold calibration can change results. Model or rubric changes can also affect comparability. Compare the questions, evidence, available checks and collection context as well as the headline score.
Do skipped checks and failed checks affect readiness in the same way?
No. Skipped checks are excluded from their category’s calculation. Evaluated failures contribute their score, often zero. A readiness error recorded with positive weight and a zero score also remains in the calculation; some checks instead return a skipped result when data is unavailable. Read the recorded status and reason before treating a zero as a confirmed website issue.
Are the report language and the AI-search market the same setting?
The report language controls presentation. Visibility questions use the inferred search language and market context. Software and global brands are generally queried without an invented local market; local and national businesses can use relevant country and language targeting. Translating a report does not by itself create a new visibility measurement in another country.
Are llms.txt, MCP.json or special schema required to appear in AI answers?
They are not universal prerequisites. AI Page Ready includes optional file checks in its current rubric, but presence does not prove visibility impact. Google states that it ignores llms.txt for Search visibility and requires no special schema for its generative AI features. A missing optional file should be interpreted in the context of the systems the business actually uses.
Is this continuous monitoring of my brand?
These reports are point-in-time audits. The visibility score summarizes valid checks for the questions and engines selected for that run. Cached answers may be reused within the current daily cache window. The report does not measure all users, personalized conversations, historical trends or every AI platform.
How are the AI answers collected?
The collection route depends on query context and configuration. Location-neutral queries can use direct OpenAI and Gemini APIs. Market-targeted queries can use DataForSEO’s ChatGPT and Gemini collection services, with direct-API fallback when collection fails. These paths may produce different answers.
The production backend configuration reviewed on 7 October 2026 selects DataForSEO providers with grounding enabled. The historical Zapier sample does not retain per-answer model or collection-provider provenance, so we cannot identify its exact answering model or assert that each answer was collected from a consumer interface.
The sample report shows the stored responses and sources. Visibility was collected on 30 September 2026, 09:26–09:28 UTC; the sample-report page presents the snapshot as 1 October 2026.
A methodology should show its workings, too.
This explanation was reviewed against the scoring and collection implementation on 7 October 2026. It is an editorial version, not a stored scoring-engine version for the historical sample.
Weights, checks and collection behavior can evolve. This page records the current explanation and the dated example; older reports may not retain every setting needed for a precise comparison.