AI answer extractability checker
The AI answer extractability checker fetches your page and tests whether ChatGPT, Perplexity or a Google AI Overview could quote a clean, self-contained passage from it. It runs nine weighted structural checks, scores the page out of 100 and gives a specific fix for every check that fails.
- Checks
- 9, weighted
- Built from
- a 243-site study
- Cost
- free, no signup
Nine checks
Open any check for why it matters and how to fix it
-
Not runDirect answer near the top—– / 20
Why it mattersA model lifts a self-contained paragraph that answers the question. This is the highest-weighted check, and 79% of the top sites in our study failed it.
How to fixOpen with a 25 to 90 word paragraph that answers the page's main question in plain terms, before any slogan or carousel.
Passed by 20.6% of 243 top homepages in the study. Worth 20 of 110 raw points.
-
Not runExactly one H1—– / 10
Why it mattersOne H1 states what the page is about. In the study, 105 of 243 homepages had none.
How to fixKeep a single H1 that names the page's subject, and demote logo or banner headings to plain text.
Passed by 46.1% of 243 top homepages in the study. Worth 10 of 110 raw points.
-
Not runThree or more H2 sections—– / 15
Why it mattersLabelled sections tell a model where one answer ends and the next begins.
How to fixBreak the page into at least three H2 sections, each named for the question or topic it covers.
Passed by 58.4% of 243 top homepages in the study. Worth 15 of 110 raw points.
-
Not runQuestion-form headings—– / 10
Why it mattersA heading phrased as the question a buyer asks maps straight onto the prompt they type.
How to fixRephrase at least two headings as the questions they answer, for example "How long does setup take?"
Passed by 16.9% of 243 top homepages in the study. Worth 10 of 110 raw points.
-
Not runLists or tables present—– / 15
Why it mattersSteps, specs and comparisons in a list or table come out with their boundaries intact, so a quote does not run two items together.
How to fixTurn any run of steps, features or prices into an ordered list, or a table with a header row.
Passed by 44.9% of 243 top homepages in the study. Worth 15 of 110 raw points.
-
Not runParagraphs kept liftable—– / 10
Why it mattersLong paragraphs blur several claims together, so any quote risks distortion. 98.8% of sites in the study passed this one.
How to fixSplit any paragraph over 120 words so each holds one claim.
Passed by 98.8% of 243 top homepages in the study. Worth 10 of 110 raw points.
-
Not runStructured data present—– / 10
Why it mattersJSON-LD states facts about the organisation, product or article in a form machines parse without guessing.
How to fixAdd JSON-LD that matches the visible copy, then validate it with the Schema Markup Validator.
Passed by 53.1% of 243 top homepages in the study. Worth 10 of 110 raw points.
-
Not runAuthor and dates declared—– / 10
Why it mattersA quote needs a source and a date to be attributable. Only 5.8% of top sites declared both, the weakest result in the study.
How to fixShow a named author with published and updated dates, and mirror them in Article and Person schema.
Passed by 5.8% of 243 top homepages in the study. Worth 10 of 110 raw points.
-
Not runCites at least two external domains—– / 10
Why it mattersLinks to outside sources show the claims can be checked, which makes the passage safer to repeat.
How to fixCite primary sources for figures and definitions with ordinary links to their domains.
Passed by 44.0% of 243 top homepages in the study. Worth 10 of 110 raw points.
What the checker read
The opening paragraph, heading outline and schema types found in the HTML
Nothing to show yet. After a check, the opening paragraph the checker found, the page's heading outline and its structured data types appear here.
110 points, nine checks, one answer
The checker uses the model from our 243-site study: nine weighted checks worth 110 raw points, normalised to 100. Each check passes or fails, with no partial credit. The direct answer carries the most weight because it is the passage a model lifts first.
| Check | Raw points | Of 100 | Passes when |
|---|---|---|---|
| Direct answer near the top | 20 | 18.2 | The first paragraph of eight or more words is 25 to 90 words long. |
| Lists or tables present | 15 | 13.6 | The main content holds at least two lists or tables between them. |
| Three or more H2 sections | 15 | 13.6 | The content is split into at least three H2 sections. |
| Exactly one H1 | 10 | 9.1 | The page has one H1, not zero and not several. |
| Question-form headings | 10 | 9.1 | At least two headings start with a question word or end in a question mark. |
| Paragraphs kept liftable | 10 | 9.1 | No more than two paragraphs run over 120 words. |
| Structured data present | 10 | 9.1 | The page carries at least one JSON-LD schema.org type. |
| Author and dates declared | 10 | 9.1 | The HTML declares an author and a datePublished or dateModified value. |
| Cites at least two external domains | 10 | 9.1 | The main content links out to two or more other domains. |
A model would struggle to quote the page without distorting it.
Extractable in parts, one or two fixes short.
A liftable answer, clear boundaries, attributable.
The weights are the study author's judgement, not a published standard. The checker reads the HTML your server returns, after removing scripts, navigation, header, footer and forms. It does not run JavaScript, and it does not measure authority, backlinks or live AI citations. A high score does not guarantee a citation.
How to run and read the check
One URL at a time, about ten seconds each. Run it on the page you want cited, not your homepage.
-
Step 1
Enter the page address
Paste the full URL of an article, guide or service page. The checker fetches the live HTML from our server, so the page must be public and not blocked to automated requests.
-
Step 2
Read the score and band
The score is the share of the 110 raw points the page earns, shown out of 100. Below 55 is Weak, 55 to 79 is Workable, 80 and up is Strong.
-
Step 3
Fix the biggest gain first
The summary names the failing check worth the most points. Open it for why it matters and the exact fix, then work down the list.
-
Step 4
Check what the checker read
The details panel shows the opening paragraph, heading outline and schema types it found. If they are empty but the page looks full in a browser, the content is rendered by JavaScript and most AI fetchers cannot see it.
-
Step 5
Re-run after you publish
Make the edits, publish, and check the same URL again. Copy or download the report to share the before and after with your team.
3%of top websites are structurally quotable by AI
We scored the homepages of 243 top websites from the Tranco list with these nine checks. 8 scored Strong. The median score was 45 out of 100.
- 18 wordsMedian opening paragraph, against a target of 25 to 90. 79% of sites failed the highest-weighted check.
- 105Homepages with no H1 at all. Fewer than half had exactly one.
- 5.8%Declared both an author and dates, the weakest result in the study.
By Sakshi Kumari, SEO Specialist at MoxSEO. Published 26 July 2026.
Read the full study- Author and dates declared5.8%
- Question-form headings16.9%
- Direct answer near the top20.6%
- Cites external sources44.0%
- Lists or tables present44.9%
- Exactly one H146.1%
- Structured data present53.1%
- Three or more H2 sections58.4%
- No oversized paragraphs98.8%
Homepages, not articles, so editorial pages likely score higher. 506 domains attempted; 243 returned a fetchable homepage.
What extractability is, and what it is not
Answer extractability is how easily a language model can lift a passage from your page and repeat it accurately without rewriting it. A page with a direct opening answer, labelled sections, lists and declared authorship gives a model clean boundaries and a source to credit.
It is one input to AI visibility, not the whole of it. A model also has to be allowed to fetch the page, find it in time, and trust the site. Check crawler access with the AI Crawler Access Checker and fetch speed with the AI Fetch Timeout Risk Checker. In our AI crawler-blocking study of 506 domains, 72 of the top sites blocked GPTBot outright, and no amount of structure helps a page a crawler cannot read.
What the checker does
- Fetches the live HTML of one public URL
- Strips scripts, navigation, header, footer and forms before reading
- Scores nine structural checks with fixed weights
- Shows the opening paragraph, heading outline and schema types it found
- Exports a plain-text report you can share
What it does not do
- Run JavaScript or log in
- Track whether ChatGPT, Perplexity or Google actually cite you
- Measure authority, backlinks or brand mentions
- Judge whether your answer is correct
- Guarantee a citation for a high score
Need help prioritising structural fixes?
Bring the page you want quoted and we can review its answer structure, attribution and implementation priorities.
Questions about AI extractability
Short, standalone answers. Email editorial@moxseo.com if any of them is wrong.
What does the AI Answer Extractability Checker measure?
It measures whether a page is structured so an AI answer engine can quote a clean passage from it. It fetches the page and runs nine weighted checks: a direct answer near the top, one H1, three or more H2 sections, question-form headings, lists or tables, paragraph length, structured data, author and dates, and external citations. The result is a score out of 100.
How is the extractability score calculated?
Each of the nine checks passes or fails. Passing checks add their weight to a raw total out of 110, which is then shown out of 100. The direct answer near the top is worth 20 raw points, lists or tables and H2 sections 15 each, and the other six checks 10 each. Below 55 is Weak, 55 to 79 is Workable and 80 or more is Strong.
Is this the same as a GEO audit?
It covers the on-page structure part of generative engine optimisation (GEO). A full GEO audit also looks at whether AI crawlers can reach the site, how fast pages respond, entity and brand signals, and whether the site is actually cited in AI answers. Use this checker for structure and pair it with the other MoxSEO tools for the rest.
Why did my page score low when it looks fine in a browser?
The checker reads the HTML your server returns and does not run JavaScript. If your content is rendered by JavaScript, the checker sees an almost empty page, and so do most AI fetchers. The details panel shows what the checker found. If it is empty, server-render the main content.
Does a high score mean ChatGPT or Perplexity will cite my page?
No. A high score means the page is easy to quote, not that it will be quoted. Citation also depends on whether AI crawlers can access the site, whether the page is found for the prompt, and how much the engine trusts the source. The checker does not track live AI citations.
Which page should I check?
Check the page you want quoted, such as a guide, article, product or service page. Homepages usually score lower because they open with slogans and navigation rather than an answer. In our study of 243 top homepages, the median score was 45 out of 100.
Where do the weights come from?
They come from the scoring model in MoxSEO's answer extractability study, published 26 July 2026 by Sakshi Kumari. The weights are the study author's judgement of what matters most for quoting, not a standard published by any AI company.
Is the checker free, and is my URL stored?
Yes, it is free with no signup. The page is fetched from our server so the browser's cross-site limits do not apply. It keeps only a short-lived, anonymised request counter for rate limiting, and it does not ask for any personal data.
Run the next check
All 18 free tools- AI Crawler Access CheckerCan GPTBot, ClaudeBot and the search crawlers reach the page at all?
- Schema Markup ValidatorValidate the structured data this checker looks for.
- AI Fetch Timeout Risk CheckerFind out if the page is too slow for an AI fetch to finish.
- llms.txt Generator and ValidatorPoint language models at the pages you want read first.
- Person and Organization Schema GeneratorDeclare the author and publisher behind the page.
- On-Page and Indexability AuditHeadings, meta and indexability for the same URL.
Rather we fix it
for you?
A free 30-minute call with Sakshi Kumari or Ashish Khan, the specialists who run MoxSEO. Bring the page you just checked; they will go through the failing checks with you and name the fixes that move the score most.
A call is not an engagement. We never guarantee rankings or AI citations.