Scoring model v4.1.0
How the score works
One audit covers Google Search, AI Overviews, AI Mode, ChatGPT, Perplexity and Claude. Google says its AI features are “rooted in our core Search ranking and quality systems”, so most of this is SEO done properly. The other engines add their own crawlers, which don't run JavaScript and can be turned away by robots.txt or a CDN.
- Points available
- 141, across 4 categories
- Blocking failures
- cap the score at 39
- Checks that don't apply
- leave the score and the total alike
Evidence grades
Every check states how strong the case for it is. Most studies in this field are correlational and run by vendors; the grades say so.
- Platform docsStated by Google, Bing, OpenAI, Anthropic, Perplexity or Cloudflare in their own documentation.
- Strong evidenceA direct mechanism, or consistent results across large studies.
- Moderate evidenceSupported by several studies, with real caveats.
- Weak evidenceSmall or inconsistent effects in the studies that exist.
- UntestedPlausible, but no study has measured it as an AI-citation factor.
- Web basicsStandard web practice. Low stakes for AI visibility on its own.
- Evidence againstPlatform guidance or controlled tests say it does not work as claimed.
What we don't score — and why
The tactics most often sold as “AI optimisation” are the ones the evidence is weakest on. If an audit charges you to fix these, ask for its sources.
llms.txt
Evidence againstAcross 137,000 domains, 97% of published llms.txt files received no requests at all in a month, and a 300,000-domain study found no link with citations.
“You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search.”
Schema and FAQ markup as a citation lever
Evidence againstA controlled test of 1,885 pages found no lift in AI Overviews, AI Mode or ChatGPT, and engines read JSON-LD as plain text or not at all. Google deprecated FAQ rich results in May 2026. We still report your structured data — it matters for rich results.
“Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add.”
Chunking content into short Q&A blocks
Evidence againstOur previous model rewarded it. The largest correlational dataset found question headings and FAQ blocks slightly negative for ChatGPT citations, and longer sections cited more than very short ones. We now score only whether the page leads with a substantive answer.
“There's no requirement to break your content into tiny pieces for AI to better understand it.”
Rewriting for AI
Evidence againstProduction benchmarks found most 'generative engine optimisation' rewriting tactics flat or negative once competitors use them too.
“You don't need to write in a specific way just for generative AI search.”
Retrieval
67 points
Whether AI crawlers can fetch and read this page at all.
| Check | Points | Evidence | Why it is here |
|---|---|---|---|
| Page fetchcan cap the score | 10 | Platform docs | Google uses only publicly accessible, crawlable content, and every AI crawler has to fetch a page before it can cite it.[1] Google: optimizing for generative AI features (May 2026) |
| Indexabilitycan cap the scoreRead the research | 10 | Platform docs | To appear in AI Overviews or AI Mode a page must be indexed and eligible for a snippet, and the site must be included in Search Console's Search generative AI control — the default, but invisible from outside. noindex removes the page.[1] Google: optimizing for generative AI features (May 2026)[2] Google: block indexing with noindex[3] Google Search Console: Search generative AI control |
| robots.txt AI rulescan cap the scoreRead the research | 10 | Platform docs | OAI-SearchBot, Claude-SearchBot and PerplexityBot obey robots.txt; blocking them keeps the site out of ChatGPT search, Claude and Perplexity answers. Most user-triggered fetchers (ChatGPT-User, Perplexity-User, Google-Agent) say robots.txt may not apply to them; Claude-User honours it.[1] OpenAI: crawlers and user agents[2] Anthropic: ClaudeBot, Claude-SearchBot, Claude-User[3] Perplexity: crawlers[4] Google: user-triggered fetchers (Google-Agent)[5] RFC 9309: Robots Exclusion Protocol |
| Raw HTML visibilitycan cap the scoreRead the research | 15 | Strong evidence | GPTBot, ClaudeBot and PerplexityBot fetch JavaScript files but do not run them, so content rendered in the browser is invisible to them.[1] Vercel & MERJ: AI crawlers don't execute JavaScript |
| Snippet controlscan cap the scoreRead the research | 6 | Platform docs | nosnippet, max-snippet and data-nosnippet limit what AI Overviews and AI Mode can show; Bing's noarchive keeps content out of Copilot answers. Google's separate site-level opt-out, the Search Console generative AI control, cannot be seen from outside.[1] Google: robots meta tag (nosnippet applies to AI Overviews and AI Mode)[2] Bing: controls for content use in Copilot (noarchive / nocache)[3] Google Search Console: Search generative AI control |
| HTTPScan cap the score | 3 | Web basics | Part of Google's page experience; no study isolates it as an AI-citation factor.[1] Google: page experience |
| Soft 404can cap the score | 3 | Platform docs | Google treats a 200 page that reads as 'not found' as an error, and drops it.[1] Google: soft 404 errors |
| Edge accessRead the research | 4 | Strong evidence | A crawler a CDN turns away cannot cite the page. Cloudflare has blocked AI crawlers by default since July 2025. Our test is simulated: real bots are verified by IP.[1] Cloudflare: AI crawlers blocked by default (Jul 2025)[2] Cloudflare: Search / Agent / Training categories (Jul 2026)[3] Cloudflare: how verified bots are identified |
| Load timeRead the research | 3 | Weak evidence | Across 107k pages, slower pages were only slightly less visible (r ≈ −0.1 to −0.18). It matters at the slow extreme.[1] Core Web Vitals vs AI visibility (107k pages) |
| CanonicalRead the research | 3 | Platform docs | A canonical pointing elsewhere tells engines to credit that URL instead of this one.[1] Google: consolidate duplicate URLs |
Extractability
28 points
Whether an engine can lift a clean, quotable answer from the content.
| Check | Points | Evidence | Why it is here |
|---|---|---|---|
| Answer first | 5 | Moderate evidence | 44% of ChatGPT citations come from the first 30% of a page. Position is one of the more reproducible effects in the research.[1] Growth Memo: 44% of ChatGPT citations come from the first 30% of a page |
| Semantic HTMLRead the research | 6 | Platform docs | Google's AI guide asks for semantic HTML and a clear structure so systems can parse the page.[1] Google: optimizing for generative AI features (May 2026) |
| Heading depthRead the research | 4 | Platform docs | Google asks for content organised in sections with headings. Descriptive studies find cited pages use more headings.[1] Google: optimizing for generative AI features (May 2026)[2] AirOps: structure of pages ChatGPT cites (descriptive) |
| H1Read the research | 4 | Web basics | A single clear H1 is basic page structure; not tested as a separate factor.[1] Google: optimizing for generative AI features (May 2026) |
| Statistics density | 3 | Weak evidence | Lab tests showed gains from adding statistics, mostly for lower-ranked pages; production benchmarks found most such tactics flat.[1] Aggarwal et al., GEO (KDD 2024, simulator)[2] C-SEO Bench (NeurIPS 2025): most tactics flat in production |
| Content depth | 2 | Weak evidence | Contradictory: one large dataset favours long pages, another found focused shorter pages win in ChatGPT.[1] SE Ranking: ChatGPT citation factors (129k domains, correlational) |
| Readability | 2 | Untested | Plausible, but no study has tested reading ease as an AI-citation factor.[1] Google: optimizing for generative AI features (May 2026) |
| Title tag | 2 | Web basics | The title is the default label for a page in search results and most AI citations.[1] Google: title links |
Authority
31 points
Whether an engine has reason to trust and attribute this source.
| Check | Points | Evidence | Why it is here |
|---|---|---|---|
| Author attributionarticle onlyRead the research | 6 | Untested | Google asks for expert-led content, but no large study has isolated bylines as an AI-citation factor.[1] Google: optimizing for generative AI features (May 2026) |
| Publication datearticle only | 4 | Moderate evidence | Cited pages are fresher than organic results, and a controlled test showed dates shifting LLM source choice. Strongest for ChatGPT; no effect seen in AI Overviews.[1] Ahrefs: AI assistants cite fresher content (17M citations)[2] Controlled test: dates shift LLM source choice |
| Last modified datearticle only | 3 | Moderate evidence | Same evidence as publication dates: freshness matters most to ChatGPT.[1] Ahrefs: AI assistants cite fresher content (17M citations)[2] Controlled test: dates shift LLM source choice |
| Freshness signalshomepage, product, other only | 5 | Moderate evidence | The only factor with both a controlled test and large observational data behind it — strongest for ChatGPT, absent in AI Overviews.[1] Ahrefs: AI assistants cite fresher content (17M citations)[2] Controlled test: dates shift LLM source choice |
| About page | 4 | Untested | A proxy for 'who is behind this'. Not tested on its own.[1] Google: optimizing for generative AI features (May 2026) |
| Contact page | 3 | Untested | A proxy for a real organisation. Not tested on its own.[1] Google: optimizing for generative AI features (May 2026) |
| External links | 3 | Weak evidence | Citing sources helped in lab tests; production evidence is weak.[1] Aggarwal et al., GEO (KDD 2024, simulator)[2] C-SEO Bench (NeurIPS 2025): most tactics flat in production |
| Knowledge graph entityRead the research | 3 | Moderate evidence | Wikipedia is among the domains AI answers cite most, and off-site brand presence correlates strongly with AI visibility. The Wikidata item is a proxy for that presence.[1] Profound: which domains AI platforms cite (680M citations)[2] Ahrefs: brand mentions vs AI visibility (75k brands)[3] Wikidata: notability policy |
Hygiene
15 points
Standard web practice. Rarely decisive for AI citation.
| Check | Points | Evidence | Why it is here |
|---|---|---|---|
| sitemap.xml | 3 | Web basics | Helps crawlers discover pages; Google documents it as a discovery aid, not a ranking factor.[1] Google: sitemaps overview |
| Sitemap in robots | 2 | Web basics | Lets crawlers find the sitemap without guessing.[1] Google: sitemaps overview |
| Internal links | 3 | Web basics | How crawlers discover and contextualise related pages.[1] Google: optimizing for generative AI features (May 2026) |
| Meta description | 2 | Web basics | Often used for the snippet; not an AI-citation factor in any study.[1] Google: meta descriptions and snippets |
| Image alt text | 2 | Web basics | Google's AI guide recommends high-quality images; alt text is how they are described.[1] Google: optimizing for generative AI features (May 2026) |
| Lang attribute | 1 | Web basics | Declares the language for every parser.[1] Google: optimizing for generative AI features (May 2026) |
| Viewport | 1 | Web basics | Part of displaying well on all devices, which Google's guide asks for.[1] Google: optimizing for generative AI features (May 2026) |
| Open GraphRead the research | 1 | Web basics | Controls link previews and some citation cards.[1] Google: optimizing for generative AI features (May 2026) |
Reported, never scored
Context we show in every report but give no points — either because the evidence says it doesn't matter, or because no study can yet say how much it does.
| Check | Points | Evidence | Why it is here |
|---|---|---|---|
| Passage length | — | Evidence against | Google says content needn't be broken into small pieces for AI; the largest dataset found 120–180-word sections cited more than very short ones.[1] Google: optimizing for generative AI features (May 2026)[2] SE Ranking: ChatGPT citation factors (129k domains, correlational) |
| Entity density | — | Untested | No study has tested named-entity density as a citation factor.[1] Google: optimizing for generative AI features (May 2026) |
| Third-party profilesRead the research | — | Strong evidence | Brand mentions off-site — YouTube most of all — correlate far more with AI visibility than backlinks. A link from your page only shows the profile exists.[1] Ahrefs: brand mentions vs AI visibility (75k brands)[2] Profound: which domains AI platforms cite (680M citations) |
| AI usage signalsRead the research | — | Platform docs | Content-Signal (Cloudflare: search, ai-input, ai-train) and Content-Usage (IETF aipref draft: search, ai-use, train-ai) state how AI may use content. Since 15 Sep 2026, Cloudflare's "Disallow AI Training" keeps Googlebot, Bingbot and Applebot crawling for search; only Training set to "Block" stops them, search included.[1] Cloudflare: managed robots.txt and Content Signals[2] Cloudflare: Disallow AI Training keeps mixed-use crawlers in search (Sep 2026)[3] IETF aipref vocabulary: train-ai, ai-use, search (draft)[4] IETF aipref: Content-Usage in robots.txt (draft) |
| Webmaster verificationRead the research | — | Platform docs | Search Console's Generative AI report and Bing's AI Performance report are the first-party ways to measure AI visibility.[1] Google: Search Console Generative AI performance report[2] Bing Webmaster Tools: AI Performance report |
| Markdown for agents | — | Untested | Google says Search doesn't need Markdown; some agents and CDN features use it.[1] Google: optimizing for generative AI features (May 2026)[2] Cloudflare: Markdown for Agents |
| Agent-usable controls | — | Platform docs | web.dev's guidance for AI agents: real buttons and links, labelled fields, no covering overlays.[1] web.dev: build sites AI agents can use (Apr 2026) |
| llms.txtRead the research | — | Evidence against | 97% of published files were never requested in a month, no correlation with citations across 300k domains, and Google says Search doesn't use them.[1] Ahrefs, Jun 2026: 97% of llms.txt files got zero requests (137k domains)[2] SE Ranking: llms.txt shows no correlation (300k domains)[3] Google: optimizing for generative AI features (May 2026) |
| Schema.orgRead the research | — | Evidence against | As a citation lever: a controlled test found no lift, and engines read JSON-LD as plain text or not at all. Still useful for rich results.[1] Ahrefs: schema and AI citations (controlled, 1,885 pages)[2] searchVIU: what AI engines read from JSON-LD[3] Google: optimizing for generative AI features (May 2026) |
| FAQ schemaRead the research | — | Evidence against | No measurable citation effect, and Google deprecated FAQ rich results in May 2026.[1] Ahrefs: schema and AI citations (controlled, 1,885 pages)[2] Google Search Central: documentation updates (FAQ rich results deprecated, May 2026) |
| Person schema | — | Evidence against | Google says no special schema is needed for AI features.[1] Google: optimizing for generative AI features (May 2026) |
| HowTo schema | — | Evidence against | Google says no special schema is needed for AI features.[1] Google: optimizing for generative AI features (May 2026) |
| Speakable schema | — | Evidence against | Google says no special schema is needed for AI features.[1] Google: optimizing for generative AI features (May 2026) |
| Hreflang | — | Web basics | Matters only for multilingual sites.[1] Google: optimizing for generative AI features (May 2026) |
| Security headers | — | Web basics | Security hygiene; no effect on AI visibility.[1] Google: page experience |
| CSP strictness | — | Web basics | Security hygiene; no effect on AI visibility.[1] Google: page experience |
| SRI | — | Web basics | Security hygiene; no effect on AI visibility.[1] Google: page experience |
| robots.txt | — | Platform docs | The file every compliant crawler reads first.[1] RFC 9309: Robots Exclusion Protocol |
| Crawl delay | — | Platform docs | Anthropic honours Crawl-delay; most other AI crawlers ignore it.[1] Anthropic: ClaudeBot, Claude-SearchBot, Claude-User |
| Favicon | — | Web basics | Cosmetic in some search and AI surfaces.[1] Google: optimizing for generative AI features (May 2026) |
| Twitter Cards | — | Web basics | Social previews only; X falls back to Open Graph.[1] Google: optimizing for generative AI features (May 2026) |
| Image formats | — | Web basics | Performance hygiene.[1] Google: page experience |
| Lazy loading | — | Web basics | Performance hygiene.[1] Google: page experience |
| Shadow DOM | — | Strong evidence | Content inside a shadow root is not in the HTML that non-rendering crawlers receive.[1] Vercel & MERJ: AI crawlers don't execute JavaScript |
What a one-page scan can't see
The best-evidenced drivers of AI citations sit off your site: brand mentions across the web (YouTube most of all), presence on review sites, Reddit and “best of” lists, and where you already rank. This audit measures whether engines can reach, read and attribute your page — the part you control directly — and checks your Wikidata entity as the one off-site signal it can verify. It does not tell you whether you are cited today.
Run a free audit