Lighthouse for AI search
Check if AI search engines can cite your page
Paste a URL and CiteSnap scores it 0–100 on whether ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews can find it, parse it and quote it — then lists the fixes in the order worth doing them.
Free, no card. Sign in once and your scan runs straight away.
- 14 checks, 100 points
- 12 AI crawlers tracked
- Nothing sent anywhere by the extension
Ready to be cited
- Retrieved25
- Whether an AI crawler can fetch this URL and parse it without executing any JavaScript first.
- Ranked15
- Whether the page states who wrote it, when it was last updated, and what single subject it covers.
- Selected60
- Whether any passage on the page still answers the question once it is lifted out on its own.
That is CiteSnap’s own home page, scored by CiteSnap. Every public page here has to clear 85 or the build fails, because a tool that scores its own pages badly is not worth listening to.
The engines that answer
- ChatGPT
- Perplexity
- Google AI Overviews
- Gemini
- Claude
Methodology
What CiteSnap measures
14 checks across three levers, weighted by how much each one actually costs you a citation. The weighting is the opinion in the product: crawler access is table stakes, but whether your content survives being chunked and ranked is where almost every page loses.
| Lever | Points | What it asks |
|---|---|---|
| Retrieved | 25 | Can a crawler fetch and parse the page at all? |
| Ranked | 15 | Does the page say who wrote it, when, and about what? |
| Selected | 60 | Is there a passage an engine can lift and quote? |
Grades run A ≥ 85, B 70–84, C 55–69, D 40–54, F under 40. A check that cannot be measured — a robots.txt that times out, say — is reported as not checked and removed from the denominator, never scored as a zero. A score you cannot trust is worse than no score.
The heaviest single check is answer-first structure at 13 points, followed by chunk extractability at 10 and heading hierarchy at 9. Retrieval checks are cheaper individually — crawler access is 8 points, indexability 6 — because they are usually one-line fixes in robots.txt or a meta tag, while an answer buried under two paragraphs of preamble is a rewrite. Google’s Lighthouse now audits some of the same retrieval signals; CiteSnap keeps its depth in Lever 3, where nothing else is looking.
Failure modes
Why pages that rank well still do not get cited
Because ranking and quoting are different jobs. A search engine returns your page and lets the reader do the extraction; an answer engine has to do the extraction itself, from a chunk, without the rest of the page for context. Three failure modes account for most of it.
13points at stake
The answer is buried
An engine reads the first passage that answers the query. Setup text is not that passage, so a section that opens by restating the question gives it nothing to lift — however good the third paragraph is.
“In this guide we’ll explore…”
answer-first
10points at stake
The paragraph does not stand alone
Answers are assembled from chunks, not pages. A paragraph opening with a bare pronoun is meaningless once separated from what it referred to, so it is never the chunk that gets quoted.
“This means the opposite is also true.”
extractability
8points at stake
The page is invisible
Client-rendered content, a noindex left in from staging, or one robots.txt group blocking GPTBot. All fixable in minutes, and all fatal until they are — which is why CiteSnap flags them instead of scoring them.
User-agent: GPTBot → Disallow: /
crawler-access
The 14-point checklist
Every check, and what it costs you
Here is the whole model, weights included. Nothing is held back for the paid tier — the scoring engine is identical in the free web checker, the extension and Pro.
Retrieved
25 pts
Retrieval decides whether GPTBot, ClaudeBot, PerplexityBot and Google-Extended can fetch this URL and read it at all, without running any JavaScript.
Crawler access
8 ptcrawler-access
Reads robots.txt group by group, the way each agent does, and reports which AI crawlers are allowed on the exact path you scanned.
Render dependency
7 ptrender-dependency
Compares raw HTML against rendered text. Content that only appears after JavaScript runs is content most AI crawlers never receive.
Indexability
6 ptindexability
The noindex tag, the canonical target and the status code — three ways a page quietly removes itself from an index.
llms.txt
4 ptllms-txt
Checks for a published llms.txt. It is an emerging convention with no proven citation impact yet, which is why it is worth four points and not twenty.
Ranked
15 pts
Ranking measures whether the page names an author, carries a machine-readable date, and holds to one subject tightly enough to be trusted as a source.
Entity clarity
7 ptentity-clarity
Named author, publisher and organisation markup — whether a model can establish who is speaking and why they would know.
Freshness
4 ptfreshness
Published and modified dates, and whether they are machine-readable rather than only printed for humans.
Topical focus
4 pttopical-focus
How tightly the page holds one subject. A page about everything is retrieved for nothing.
Selected
60 pts
Selection is where almost every page that already ranks on Google loses the citation, because a passage has to survive being lifted out on its own.
Answer-first structure
13 ptanswer-first
Whether each section answers its own heading in the first sentence. This is the heaviest check on the model because it is the one that decides whether there is anything to lift.
Chunk extractability
10 ptextractability
Whether a paragraph still means something once it is lifted out of the page — which is the only state an answer engine ever reads it in.
Heading hierarchy
9 ptheadings
Whether the outline is well-formed and whether headings read as the questions their sections answer.
Scannability
9 ptscannability
Paragraph length, sentence length, lists and tables — the structure that survives being split into chunks.
Specifics
8 ptspecifics
Numbers, dates, names and units. Concrete claims get quoted; generalities get paraphrased into someone else’s answer.
Question and answer pairs
7 ptfaq
Detects genuine Q&A structure, the format answer engines lift most readily because the question is already matched for them.
Schema
4 ptschema
Validates JSON-LD for the four types that carry citation weight: Article, FAQPage, HowTo and Organization.
Chrome extension
Scan the page you are already on
The extension scores the page in front of you, including staging sites and pages behind a login that this server cannot reach. It runs entirely in your browser: the Manifest V3 build requests exactly two permissions, activeTab and scripting, so Chrome shows no site-access warning at install and the scanner runs only after you click Scan. It scores what it finds with the same @citesnap/engine package this server runs, so a page cannot pass in one surface and fail in the other.
- 1Open CiteSnap in the Chrome Web Store and add it.
- 2Chrome asks for no site access, because the manifest requests none.
- 3Click Scan on any page — including staging and pages behind a login.
- Install-time site access
- None
- Data sent to CiteSnap
- None
- Scoring engine
- Identical
Crawlers
The crawlers it checks by name
CiteSnap reads your robots.txt the way each of these agents does, group by group, and tells you which are allowed on the path you scanned. These four carry the most weight, because they build the indexes the big answer engines actually search.
GPTBot
OpenAI (training + search index)
OAI-SearchBot
ChatGPT Search
ClaudeBot
Anthropic / Claude
PerplexityBot
Perplexity index
Google-Extended
Google AI Overviews / Gemini
- ChatGPT-User
- anthropic-ai
- Perplexity-User
- CCBot
- Bytespider
- Amazonbot
- Applebot-Extended
How to use it
Start with the GEO score checker on any page you already care about ranking. It takes about five seconds and needs no account.
- Scan the page and read the blocking issues first, if there are any.
- Work the fix list top-down — it is ordered by points recovered, then by effort.
- Copy the fix prompt and hand it to a coding agent; it carries the evidence behind every check, not just the verdict.
- Re-scan to confirm the change landed.
The four free tools
All four tools work logged out, with no trial clock, no credit card and no cap on how many pages you run through them.
Tool 01
GEO score checker
Score any URL 0–100 on AI citation readiness and get the prioritized fixes, in about five seconds.
Tool 02
AI crawler checker
Read a site’s robots.txt and show exactly which AI crawlers — GPTBot, ClaudeBot, PerplexityBot, Google-Extended — are allowed or blocked.
Tool 03
llms.txt generator
Generate a spec-compliant llms.txt from a site’s sitemap, or validate the one already published against the spec.
Tool 04
Schema generator
Build valid FAQPage, Article, HowTo or Organization JSON-LD tuned for citation, entirely in the browser.
Pricing
Free to scan. Pro to keep it that way.
Scoring a page costs nothing and always will — no account, no card, no trial clock. Pro is $12 a month and answers the harder question: it stayed fixed, right?
Free
$0forever
The whole scoring engine, on every page you point it at, with no account and no monthly limit.
- Unlimited page scans, on the web and in the extension
- All four GEO tools, no account needed
- Prioritized fixes and the copyable agent prompt
- Shareable report card
- Saved history between scans
Most popular
Pro
$12/ month, or $99 a year
Pro adds persistence and depth, for people who publish continuously and need a score to stay up rather than go up once.
- Everything in Free
- Saved history and trend lines per URL
- Bulk audit up to 100 pages from a sitemap
- Daily monitoring with email alerts on a drop
- Inline fixes in your editor as you write
- AI checks on your own API key
Questions
Frequently asked questions
What is a GEO score?
A GEO score is a 0–100 measure of how ready a page is to be cited by an AI answer engine. CiteSnap computes it from 14 checks across three levers: whether a crawler can retrieve the page (25 points), whether the source reads as trustworthy and current (15 points), and whether the content is extractable and quotable (60 points).
How is GEO different from SEO?
SEO optimises for a ranked list of links; GEO optimises for being quoted inside a generated answer. The overlap is real but partial: a page can rank first on Google and still be unusable to ChatGPT because its answer is buried under three paragraphs of setup, or because robots.txt blocks GPTBot.
Does blocking GPTBot in robots.txt stop ChatGPT citing me?
Yes, for the indexed path. GPTBot builds the index ChatGPT searches, so a site that disallows it will not appear as a source in ChatGPT answers regardless of how good the content is. CiteSnap flags this as a blocking issue rather than a scoring deduction, because no amount of content work compensates for it.
Is the CiteSnap scan free?
Yes. The web checker and all four tools work without an account, and the Chrome extension is free and runs entirely in your browser — the pages you scan with it never leave your machine.
How much does CiteSnap cost?
Scanning is free with no account and no limit. CiteSnap Pro is $12 a month, or $99 a year, and adds saved score history, bulk audits of up to 100 pages from a sitemap, daily monitoring with email alerts, inline fixes in your editor, and AI checks that run on your own API key.
Does CiteSnap run JavaScript when it scans a page?
No. The server fetches your HTML once, as an unauthenticated crawler, and never executes the page’s JavaScript. That is deliberate: most AI crawlers do not execute JavaScript either, so what CiteSnap parses is what they actually see.
By the CiteSnap team · Published · Last updated
Find out what an answer engine sees
One URL, about five seconds, no account. You get a number, the 14 checks behind it, and the fixes in the order worth doing them.