Real GEO Work vs Checklist Noise: What AI Readiness Scans Actually Measure
AI readiness scans measure schema presence, not crawler access. A three-tier triage sorts real citation blockers from checklist noise like BreadcrumbList on root pages.
Last updated:
14 min read
Summary
- AI readiness scans check whether markup types are present, not whether an AI crawler can reach your content — so a flag like "missing BreadcrumbList" on a root page measures template coverage, not readiness.
- The only hard blocker is rendering: GPTBot and ClaudeBot execute no JavaScript, so React, Vite, and client-only Next.js pages hand AI crawlers an empty
<div id="root">. - The popular fixes don't move answers: an Ahrefs test of 1,885 pages found no citation uplift from added JSON-LD, and 97% of published
llms.txtfiles received zero crawler requests in May 2026. - Spend minutes, not budget: confirm the raw HTML contains your headline and claims via
curl, checkrobots.txtandnoindexfor AI user agents, addsameAslinks, and fix inaccurate core schema — then decline breadcrumbs, completeness scores, andllms.txt. - Citations follow third-party authority over on-page artifacts, with off-site sources earning 36.3% of AI citations versus 4.6% for brands' own sites; to map those gaps across engines, Mole's AI SEO & GEO Agent tracks which prompts cite or omit you.
A B2B SaaS founder recently posted a question plenty of other people are quietly asking. They had run an AI search readiness scan on their single-page landing page. The report came back with 13 passes and three urgent fixes:
- No
BreadcrumbListorItemList— "marking up breadcrumbs and lists helps convey the page's position within the site." - No
sameAson theOrganizationschema — "helps engines confirm it's the same entity." - No visible last-updated date — "lets readers judge freshness."
The page already had Organization, WebSite, SoftwareApplication, VideoObject, and FAQPage JSON-LD. It had a sitemap with lastmod. The scanner flagged breadcrumbs on a root page with no hierarchy — a page that, by definition, cannot have a breadcrumb trail. That detail alone tells you what you are dealing with: a generic template applied without reading the page.
This isn't an isolated case. Most AI readiness scan reports work this way: they check whether a markup type exists, not whether an AI crawler can reach the content at all.
The three-tier framework below lets you sort any scan result without becoming a technical SEO.
Tier 1: Fix These First — They Are the Only Things That Genuinely Block Citation
Can an AI crawler render your page at all?
This is the question most AI readiness scan tools don't ask, and it's the only one that produces a hard zero.
Many SaaS landing pages are built with React, Next.js in client-only mode, or Vite. The server sends a near-empty HTML file — a <div id="root"> and a bundle of JavaScript. The browser executes that JavaScript and builds the visible page in memory. A human using a browser sees the full page. A crawler that does not execute JavaScript sees nothing worth citing.
A joint Vercel and MERJ analysis of more than 500 million GPTBot requests found zero evidence of JavaScript execution. GPTBot fetched .js files in roughly 11.5% of requests but never ran them. ClaudeBot downloaded JavaScript in approximately 23.8% of cases, also without execution. AI crawlers fetch the raw HTML the server returns before any JavaScript runs. That is the content they index.
Googlebot is an exception. It maintains a rendering pipeline inherited from its traditional search operation. Crawlers built from scratch for AI purposes skip rendering because executing JavaScript at scale is computationally expensive. Passing Google Search Console's coverage check does not confirm that ChatGPT or Perplexity can read your page.
How to check this yourself:
Open a terminal and run:
curl -A "Mozilla/5.0 (compatible; GPTBot/1.0)" YOUR_URL_HERE
Read the output. If your main headline, product description, and key claims appear as plain text, the page is readable. If you see an empty <div id="root">, a loading message, or nothing but <script> tags, AI crawlers see an empty page regardless of how much schema you have added.
For pages with a JavaScript rendering problem, the fix is server-side rendering or static site generation — not a schema addition. If you render the page correctly and it still takes weeks to surface, the bottleneck has moved to indexing, and getting new pages crawled and indexed quickly is the next lever rather than another markup change.
Is your page crawlable and indexable?
Before any content question arises, the crawler has to be allowed through. Check your robots.txt file and confirm you are not blocking the crawlers that matter: GPTBot (OpenAI training), OAI-SearchBot (ChatGPT search retrieval), ClaudeBot, and PerplexityBot. These are distinct user agents with separate functions. Blocking GPTBot does not remove you from ChatGPT search results, but blocking OAI-SearchBot does.
A noindex directive or an unresolved redirect chain has the same effect. A piece that isn't indexed can't be cited — the crawl gate always comes before the content question.
Does the page answer a question an AI would use?
This isn't a technical check — it's a content check. An AI engine cites a page because the page gives a useful, specific answer to a query. If the page is a vague brand statement with no factual claims, no pricing context, and no comparison information, there is nothing to extract and cite. The markup is irrelevant at that point.
Tier 2: Handle These Quickly — Minutes, Not Budget
Once the Tier 1 blockers are handled, three items from a typical GEO checklist deserve a short session. None will move the needle on their own, but all are cheap and defensible.
Add sameAs links to your Organization schema
The scanner in the opening scenario flagged this correctly, even if it misjudged the urgency. Adding sameAs links — pointing to your LinkedIn company page, Crunchbase profile, and official X account — helps AI models tie your domain to a known entity. It removes ambiguity when a model is deciding whether the "Acme" in its training data is the same company as the one at acme.io. Five minutes if you have established profiles.
Use a visible dateModified where the content is genuinely time-sensitive
The scanner's tourism example was misapplied to a SaaS page, but the underlying principle is not wrong everywhere. For pricing pages, policy documents, and articles covering fast-moving topics, a visible publication and update date is a useful signal for human readers. The dateModified field in your schema reinforces this for machines.
For a stable product landing page, this is a trust feature for humans rather than a meaningful AI ranking input. Add it where it's honest. Don't add it to convey freshness that doesn't exist.
Confirm your core schema is accurate, not complete
Forget completeness scores. The goal is accuracy. If your Organization schema states the wrong founding year, links to a dead URL, or lists a category that does not match what your business does, that is worth correcting. The same applies to SoftwareApplication markup for SaaS products. Accurate, unambiguous claims about your business help a model state facts about you without hedging.
Tier 3: These Items Do Not Warrant Your Time or Budget
Scan reports consistently mishandle this tier. The items below are either semantically wrong for the page type, measured incorrectly, or backed by no credible evidence.
BreadcrumbList on a single-page site or root page
A breadcrumb trail represents a hierarchy: Home > Category > Page. On a single-page site, or on the root page of any site, that hierarchy does not exist. Marking up a trail that does not exist produces no signal for a machine. It isn't a missing optimisation. It's a generic check that doesn't apply to this page structure.
Schema completeness as a percentage score
Any tool that reports schema coverage as a percentage is measuring presence, not effect. Schema types are either present or absent. Presence does not equal citation.
Ahrefs ran a controlled study on 1,885 pages that added JSON-LD and measured the citation effect across three surfaces:
- AI Overviews: -4.6%
- AI Mode: +2.4% (statistically indistinguishable from zero)
- ChatGPT: +2.2% (statistically indistinguishable from zero)
Schema is more common on cited pages than uncited ones — but that correlation disappears in a matched test. Adding schema types does not move AI answers. A scan that counts schema blocks and calls the result a readiness score is counting an input with no measured output.
llms.txt
The is llms.txt worth it question has a clear answer at this point: no, not as an AI visibility lever.
Google's own guidance, updated in July 2026, tells site owners to avoid "unnecessary AI text files (like llms.txt)." Google's John Mueller described the benefit as "purely speculative for now." No major AI provider — OpenAI, Anthropic, or Google — has committed to using it for citation purposes.
The adoption data confirms this. An Ahrefs study of 137,210 domains found 97% of published llms.txt files received zero crawler requests in May 2026. An SE Ranking analysis of approximately 300,000 domains found no statistically significant correlation between having an llms.txt file and AI citation frequency. Their predictive model became more accurate when llms.txt presence was removed as a factor entirely.
llms.txt is not a lever. Spending money to produce or optimise one is spending money on a file that the major platforms have confirmed they do not use. If a scan flags it as urgent, that flag tells you more about the scan than about your site.
Getting Cited Depends on Authority, Not Artifacts
The teardown above leaves a practical question open: if schema completeness and llms.txt do not move AI answers, what does?
The answer sits upstream of any on-page technical work. AI engines retrieve content from indexes that weight authority signals — and the dominant signal is third-party reference. Being mentioned, linked to, and discussed by sources other than your own domain is what places you in the pool of content an AI engine considers worth citing.
An Ahrefs study found that branded web mentions correlated with AI Overview visibility at 0.664, while backlinks correlated at 0.218. Brands' own websites earn approximately 4.6% of citations in AI answers. Third-party sites — comparison pages, review platforms, industry publications, forum threads — account for 36.3%.
The page type also matters. A 2026 study of AI citation patterns found the following median citation rates by page type:
- Discussion pages: 1.86
- Articles: 1.82
- Comparison pages: 1.70
- Homepages: 1.06
Comparison pages earned a citation rate approximately 1.6 times higher than homepages. A landing page with no question-and-answer structure, no comparison data, and no third-party validation is a poor candidate for citation regardless of its markup.
Identify the gaps before building more content
The practical sequence is:
-
List the questions your buyers actually ask. Pull from sales calls, support tickets, and the search queries that bring people to your site.
-
Check who gets cited for each question across engines. Different AI engines maintain separate indexes, so a strong Google ranking does not transfer to Perplexity or ChatGPT. The 2026 AI Search Visibility Study found a median 5.1x gap between a brand's best and worst engine for visibility. between a brand's best and worst engine for visibility. A brand invisible on Perplexity may be well-cited on Google AI Overviews — or vice versa.
-
Build the page that deserves to be cited. That means a page with specific, accurate, substantive content that answers the question better than the current cited sources.
To run that analysis at scale — tracking which prompts mention you, on which engines, and where the gaps are — Mole's AI Visibility Tracker shows where AI answers cite or omit you for a given prompt set across engines, giving a stable baseline to measure against.
Use the Scan as a Starting Point, Not a Work Order
An AI readiness scan is a useful prompt. It isn't an authoritative diagnosis, and you shouldn't treat it as a prioritised work order.
When a scan flags three urgent items and your page already carries five schema types and a sitemap with lastmod, the right response is not to open a development ticket. It is to read the flags through the lens of this framework:
Tier 1 — Fix these, they block everything:
- Can an AI crawler render the page? Check with a raw
curlrequest. - Is the page crawlable and indexable? Check
robots.txtandnoindexdirectives. - Does the page answer a question worth citing?
Tier 2 — Spend minutes, not budget:
- Add
sameAstoOrganizationschema if you have established profiles. - Add a visible
dateModifiedwhere the content is genuinely time-sensitive. - Correct inaccurate core schema claims.
Tier 3 — Decline these confidently:
BreadcrumbListon a page with no hierarchy.- Schema completeness percentages.
llms.txt.
The work that moves AI search readiness from theoretical to real is upstream: building pages that answer specific questions, earning third-party mentions, and ensuring the crawler can actually read what you have built. A scan that skips the render check and flags a missing breadcrumb trail on a root page is not measuring readiness. It is measuring template coverage.
Founders asking is GEO worth it will find the answer is yes — but only when the effort goes into the work that changes what an AI engine can read, retrieve, and cite. Scan reports that do not distinguish between those outcomes are a starting point that needs interpretation, not a list that needs completing.
Frequently Asked Questions
What is an AI search readiness scan?
An AI search readiness scan is an automated audit that checks whether specific markup types, files, and metadata are present on your site — not whether an AI crawler can actually reach your content. That distinction matters, because a page can pass every markup check and still be invisible in ChatGPT or Perplexity if its content only renders in the browser. Treat a scan report as a list of things to evaluate, never as a diagnosis.
Do you need schema for AI Overviews?
Schema is more common on cited pages than uncited ones, but Ahrefs' controlled study of 1,885 pages found no citation uplift from adding JSON-LD. Existing schema being accurate matters. Adding schema types for completeness does not demonstrably improve citation rates.
Can AI crawlers read my website if it uses React or Vite?
Not reliably. AI crawlers including GPTBot and ClaudeBot do not execute JavaScript. If your page builds its content client-side, crawlers see an empty shell. Run a curl request as shown above to confirm what the raw HTML contains.
Which AI crawlers should I allow in robots.txt?
Allow GPTBot, OAI-SearchBot, ClaudeBot, and PerplexityBot unless you have a specific reason to block one. They do different jobs: GPTBot is OpenAI's training crawler, while OAI-SearchBot handles ChatGPT search retrieval — blocking the first does not remove you from ChatGPT results, but blocking the second does. Then confirm that noindex tags and unresolved redirect chains are not quietly shutting those crawlers out, because the crawl gate always comes before the content question.
Is llms.txt worth it for AI visibility?
No current evidence supports it as an AI citation lever. Google's guidance tells site owners to ignore unnecessary AI files like it, and 97% of published files received zero crawler requests in May 2026.
When does a visible last-updated date actually matter for AI search?
A visible update date matters on pages whose accuracy changes over time — pricing, policies, and fast-moving topics — and is close to irrelevant on a stable product landing page. The dateModified field reinforces that date for machines, but the visible version is a trust signal for human readers first. Add it where it's honest, and don't manufacture freshness that doesn't exist.
What actually determines whether an AI engine cites my page?
Third-party authority is the dominant signal — mentions on comparison pages, review sites, forum threads, and industry publications. Page type also matters: articles, discussion pages, and comparison pages are cited significantly more often than homepages. Content that answers a specific question on a page designed for that purpose outperforms a polished homepage with generic brand copy.
Why did my AI readiness scan flag BreadcrumbList as urgent on my homepage?
Because the scanner is running a generic template. BreadcrumbList marks up a navigational hierarchy. A root page with no parent pages has no hierarchy to mark up. The flag reflects template coverage, not a genuine gap in your content's readability.
How do I improve my AI visibility in ChatGPT and Perplexity?
Build pages that answer specific buyer questions, earn third-party mentions, and make sure crawlers can read the raw HTML — in that order. Third-party sources account for 36.3% of AI citations versus 4.6% for brands' own sites, so off-site authority does far more work than any on-page artifact. Because each engine maintains a separate index, the fastest way to decide where to start is to see which prompts cite you, on which engine, and where the gaps sit — Mole's AI SEO & GEO Agent tracks exactly that for a defined prompt set.