Free tool · AI visibility
Free AI answer preview
To see the answer an engine can build from a page, read the elements it draws on in order: structured answers, the meta description, the opening paragraph, then the h1 and title. This assembles that answer by extraction, lifting every sentence from your page verbatim and labelling it with the element it came from, so nothing here is written by us or by a model.
Free and unlimited. The extraction preview needs no account; the model comparison needs a free sign-in and is limited to one run a month, because it costs us money.
AEO depth layer
See the Model comparison
The Model comparison is computed from your own page and costs nothing to run. We ask for a free account because the expensive tools on this shelf stay free, and because it lets us tell you when your result changes. No card, no trial clock.
- A model is given your page and nothing else, then asked what it would answer with. The summary itself is not the point: any assistant can summarise a page.
- Every sentence it produced is checked back against your page, and the ones that cannot be traced to anything you wrote are flagged.
- Those are the claims a reader would attribute to you that your page never made, which is the failure the free extraction cannot show you on its own.
Signing in saves the details you entered above so we can follow up about your results. See our privacy policy for what we keep and why. The AI Answer Preview result you already have stays free either way. Privacy policy.
The five source elements, in the order they are preferred
An engine assembling an answer does not read your page top to bottom and pick fairly. It prefers text that was written to be quoted, and falls back through progressively worse options. This is the order this tool uses, and the reason for each place in it.
| Order | Element | Why it sits here |
|---|---|---|
| 1 | Structured answers (FAQPage, described entity) | A sentence you wrote specifically to be quoted, in a format that says so. The most reliably lifted text on any page. |
| 2 | Meta description | The summary you chose. When it is missing, the engine writes its own from the body and you never see it. |
| 3 | Opening paragraph | The default extraction target. Short enough and it gets lifted whole; long enough and the engine cuts it, or quotes someone else. |
| 4 | h1 | A label, not an answer. Used to name the subject when nothing above exists. |
| 5 | Title tag | The weakest of the five for this purpose, because it is written for a result listing rather than to answer anything. |
Why the free half uses extraction rather than a model
This is the design decision the whole tool rests on, so it is worth stating rather than hiding behind the word AI.
| Property | Extraction (free) | A model call (signed in) |
|---|---|---|
| Same page, same answer | Always | No, it samples |
| Can invent a claim your page never made | No, by construction | Yes, and that is the finding |
| Shows which element each sentence came from | Yes | No |
| Costs money to run | No | Yes, so it is metered |
| Tells you what an engine actually said | No | No |
| Useful for spotting an ungrounded claim | As the baseline to compare against | Yes, that is its whole job here |
Neither half tells you what an engine actually said, because nothing that reads your HTML can. For that, the AI visibility checker asks the engines directly. To make the page more quotable in the first place, the AEO readiness score grades what an engine can extract, and the citation schema generator builds the structured answers that sit first in the order above.
How to preview the answer an engine would give from your page
Step 1
Enter a page that answers a question
Not the homepage. A page that answers something specific is the only kind an assistant has a reason to quote.
Step 2
Read the source labels, not just the answer
An answer built entirely from your title and h1 means the engine had no real sentence to work with and fell back to your labels.
Step 3
Close the gaps that force an engine to improvise
A missing meta description or a missing structured answer is the difference between an engine quoting a sentence you chose and one it extracted for itself.
Step 4
Then compare against what a model actually writes
Sign in and the deeper layer gives a model your page and nothing else, then flags every sentence in its summary that your page does not support.
When you outgrow this tool
A preview shows what one page makes available to be quoted. Answer Radar asks the engines your buyers' real questions and reports whether you are named in the answer at all.
Previewing one page is free and unlimited. Linkeddit is what you use when the question stops being what your page could be quoted for and becomes which questions your buyers ask that nothing of yours answers.
FAQ
AI Answer Preview questions
What extraction can tell you that a model cannot, and the question this tool deliberately does not claim to answer.
Is this what ChatGPT will actually say about my page?
No, and nothing that reads your HTML can tell you that. This is what an engine COULD assemble from the elements you control, in the order those elements are usually preferred. It is a prediction about your page, not a record of any engine's behaviour. We would rather state that than sell you a screenshot of something we generated and called an answer.
Why is the free version not powered by AI?
Because a model is the wrong instrument for this question. The point is to show you what your page makes available, and a model would blur that by filling gaps with things it happens to know. Extraction cannot do that: every sentence here appears on your page verbatim, which is checked by a test rather than promised in copy. Determinism is the other half. A model gives you one sample; this gives you a property of the page.
How long is the preview?
Capped at 80 words, which is the order of length an answer engine actually produces. If your page cannot fill that from the elements above, the preview is shorter and the gaps section says which elements were missing. A short preview is a finding, not a failure of the tool.
What counts as a source element?
Five, in the order an engine tends to prefer them. Structured answers first, meaning an FAQPage acceptedAnswer or a described entity in your JSON-LD, because those are sentences written specifically to be quoted. Then the meta description, which is the summary you chose. Then the opening paragraph. Then the h1 and the title, which are labels rather than answers and are used last, only to name the subject when nothing better exists.
My preview is built entirely from my title and h1. What does that mean?
That your page has no quotable sentence. An engine reaching that state either assembles an answer from fragments, which reads badly and gets attributed to you anyway, or quotes a competitor who wrote one. The fix is usually one paragraph: a direct answer in the opening, or an FAQPage block with the two questions your buyers actually ask.
What is in the layer that needs an account?
One model call, given your page and nothing else, and the comparison against the extraction above. The summary itself is not the point, since any assistant can summarise a page. The finding is the sentences the model produced that cannot be traced back to your page: given only your words it still made claims you never wrote, and a reader would attribute them to you. It costs us money, so it is one run per account per calendar month.
Why meter the model half at one run a month?
Because it is the only part that costs anything, and putting a limit on the whole tool to make the limit feel like a feature seemed worse than metering the expensive part and leaving everything else open. The extraction above is free and unlimited, including for signed-out visitors.
More free tools
Every tool on the shelf is free, with no card and no trial. Most need no account at all. See all free tools.
Build the JSON-LD properties that make a page attributable rather than eligible for a rich result, with the reason each one matters.
Score a page on whether an assistant could actually answer with it, across four pillars worth a hundred points.
Find out what is stopping a page being indexed, which is the half of the question anything can actually answer.
Whether each answer engine can reach and quote one specific page, checked per engine against the bots it actually uses.