Answermend

Free tools / ai-content-checker

Free tool · runs on our server · no engine calls

AI Content Checker

An assistant that quotes a page has to be able to lift an answer out of it: find the question, find the answer under it, and take a passage that still makes sense on its own. This reads your page and tells you where that breaks — with the headings and sentences that caused each verdict.

free · no signup · no card · queries no AI engine

Check a page

live check: off

One page, not a whole site. Shape is a property of a page — a guide and a product page on the same site can score nothing like each other.

The address you type goes to our server, which fetches the page. Your browser is not allowed to read a page from another website — that is a rule browsers enforce, not a choice we made — so there is no in-browser version of this and there cannot be. What we fetch is the page as your server delivers it, with none of its scripts run. We do not send the page, or the address, to any AI engine.

Current state This reads the page you give it, on our server. The address you submit is sent to api.answermend.com, which fetches that page and measures its shape. It queries no AI engine and spends nothing, so it is not a prediction of what ChatGPT or Gemini would say about you — it reports how the page is built. A labelled example is shown below so you can see the shape of a result before running one.

Answer shape · per check, with evidence

https://demo-clinic.example/services/teeth-whitening EXAMPLE DATA

45%

Not answer-shaped. 40.3 points earned of the 90 points this page could be measured on. 10 points of checks did not run and are left out of that total rather than counted as failures.

40.3 ÷ 90 = 45% · bands: 80%+ answer-shaped · 55–79% partly · under 55% not answer-shaped · below 40 measurable points we name no band at all

question headings13.9 of 25

2 of 9 subheadings (h2–h4) are phrased as a question — 22%. Our threshold for full points is 40%.

  • question-shaped · h2How long does teeth whitening last?
  • not question-shaped · h2Our Approach
  • not question-shaped · h2Why Choose Us
  • not question-shaped · h2Book Today
answer follows the question12.5 of 25

1 of 2 question-shaped headings is followed directly by a paragraph, list item or quote whose first sentence is between 6 and 45 words and does not begin with a warm-up phrase.

  • answeredHow long does teeth whitening last?a 14-word sentence follows the question directly — “Most treatments hold their colour for between six months and two years.”
  • delayedIs teeth whitening safe?the block after the question opens with scene-setting: “When it comes to”
paragraphs survive being quoted10.9 of 20

6 of 11 paragraphs of 25+ words start with their own subject and refer to nothing outside themselves.

  • would not survive being quoted aloneThis means you can expect a noticeably brighter shade within one appointment, though the exact result depends on…opens with a demonstrative with no noun (“This means”) — a reader who sees only this paragraph does not know what it refers to
  • would not survive being quoted aloneThe table above shows the price for each of the three whitening options we offer, including the take-home…points at another part of the page: “The table above” — true on the page, meaningless in a quote
  • self-containedTeeth whitening lightens the natural colour of enamel using a peroxide gel applied under supervision…no dependent opener and no reference to another part of the page
lists and tables0 of 10

720 words of body text and not one list or table. Everything on this page is prose, so a reader looking for “the three options” or “the price by treatment” has to extract them from sentences. This is a measured zero — the page was read and the structures are absent.

  • counted0 lists of 3+ items, 0 lists of 1–2 items, 0 tables of 2+ rows, over 720 words of body text
direct openingNOT MEASURED 10 points, unrun

There is no h1 in this page’s HTML, so there is no opening to locate. This check did not run. It is not a score of zero, and its 10 points are not in the total above.

  • unrunthe missing h1 is worth fixing, but it is a different check on a different tool
patterns that resist quoting3 of 10

Starts at 10 and loses points for each pattern found: walls of text (−4), very long sentences (−3), label headings (−3). Every deduction, and every one not taken, is listed with the count that decided it.

  • −4 walls of text3 of 11 paragraphs (27%) run past 120 words; our threshold is 15%
  • −3 very long sentences6 sentences run past 45 words; our threshold is more than 2
  • no deduction — label headings2 of 9 subheadings (22%) are short labels rather than questions, at or under our 30% threshold
Signal · counted, not scored The most repeated two-word phrase is “teeth whitening”, 19 times, 5% of body words — under our 10% mark. Not scored either way. Signals never move the points, because the points have to be reproducible from the six checks and nothing else.

Read from the page as delivered, with no JavaScript run. This is the shape of the page. It is not a claim about what any AI engine would answer, quote or cite — nothing here queried one.

Where this stands The six checks and their weights, in the page as it arrives A labelled example result Fetching your page and reading it back

Why this tool exists

A page can say the right thing and still be impossible to quote.

the same fact, two shapesEXAMPLE DATA
<!-- hard to lift an answer from -->
<h2>Our Approach</h2>
<p>When it comes to whitening, there
are many factors to consider. It
depends on the case…</p>

<!-- an answer can be lifted whole -->
<h2>How long does whitening last?</h2>
<p>Whitening holds its colour for six
months to two years, depending on
diet and smoking.</p>

An assistant answering a question does not read your page the way a visitor does. It looks for the part that answers the question, and it takes a passage. Whether that is easy or impossible is decided by how the page is built, not by how good the writing is.

Three things break it most often. A heading that names a topic instead of asking a question gives nothing to match against. An answer that arrives three paragraphs after the heading is not attached to it. And a paragraph opening with “This means…” or “The table above shows…” falls apart the moment it is separated from what came before it — which is exactly what quoting does.

This tool reads for those, and names them. Not a number with an arrow next to it: the actual heading, the actual sentence, and the rule it broke, so you can disagree with the verdict on the evidence.

It reads one page at a time, because shape is a property of a page. A guide and a booking page on the same site will not resemble each other here, and averaging them would hide the one you need to fix.

The arithmetic, printed

Six checks. Here is every weight
and every threshold.

A number you cannot rebuild from the checks under it is a number you have to take on trust. These are ours, and they are conventions we chose — not measurements of how any engine behaves.

What each check reads, what it is worth, and the line it draws
CheckWeightWhat it reads, and where the line is
Question headings25 What share of your h2–h4 headings are phrased as a question. A heading counts if it has a question mark, or opens with what, why, how, when, where, who, which, can, do, does, is, are, should, will, would or if. Full points at 40% or more; the share scales linearly below that. Fewer than two subheadings and there is no share to report, so the check does not run.
Answer follows25 For each question-shaped heading: is the very next block a paragraph, list item or quote whose first sentence runs 6–45 words and does not begin with a warm-up? Score is the share that pass. Every failure is listed with its reason — another heading, a code block, a stub, a buried lead, or a delaying opener.
Paragraphs survive quoting20 Of the paragraphs of 25 words or more, what share start with their own subject and refer to nothing outside themselves. A bare pronoun, a demonstrative with no noun (“This means…”), a discourse connective (“However,…”) or a pointer at the page (“the table above”, “as mentioned earlier”) fails it. Under three qualifying paragraphs, the check does not run.
Lists and tables10 Full points for at least one list of three or more items, or one table of two or more rows. Half points if the only lists have one or two items. Zero if a page of 400+ words has neither. Under 400 words the absence means nothing, so the check does not run instead of penalising a short page.
Direct opening10 Is the first block under the h1 a sentence of 6–45 words with no warm-up in front of it. Half points for a sentence too short to be an answer, zero for a second heading, a warm-up or a 46-word opener. No h1, or nothing after it, and the check does not run.
Patterns that resist quoting10 Starts at 10. Minus 4 if more than 15% of paragraphs run past 120 words. Minus 3 if more than two sentences run past 45 words. Minus 3 if more than 30% of subheadings are two words or fewer and are not questions. Deductions taken and not taken are both shown with their counts.
  1. The total is the sum of the six, and nothing else. No hidden term, no adjustment, no rounding you cannot see.
  2. A check that could not run leaves the total on both sides. If we could not read your paragraphs, that check’s 20 points come out of what you earned and out of what you were measured against. It is never scored zero, because “we could not look” and “we looked and found nothing” are different findings and only one of them is about your page.
  3. Below 40 measurable points we name no verdict at all. A page built out of layout tables with no headings and no paragraphs leaves almost nothing to read, and calling that page “not answer-shaped” on a tenth of the evidence would be a guess wearing a percentage.
  4. Signals are counted and never scored. Phrase repetition is reported next to the result and moves no points, because the moment a soft judgement enters the total, the total stops being reproducible.

Reading the output

How to read the result

Four rules. The second is the one that decides whether this tool is worth anything to you.

  1. Read the evidence, not the percentage. Every check lists the headings and sentences that produced it. If you disagree with a verdict, the sentence that caused it is right there and you are probably right.
  2. Fix the named sentence, not the score. “Answer follows: 1 of 2” is worth nothing on its own. “The paragraph under Is teeth whitening safe? opens with When it comes to” is a five-minute edit.
  3. Anything marked NOT MEASURED is not a pass and not a zero — it means the check had nothing to read, and the reason is printed next to it.
  4. A good score is not a promise of anything. Shape removes an obstacle. It does not create a mention, and no result on this page is a prediction about an AI engine.

What this will not tell you

Limitations, stated plainly.

Four, and the first one is the ceiling of the whole method.

Shape is not substance, and we proved it against ourselves

We wrote a page whose every heading was a question and whose every answer restated that question in different words — a page about nothing. It scored 97% of its measurable points and came back “answer-shaped”, ahead of every real page we tested. Structure is all this reads. It cannot tell a good answer from a well-formatted tautology, and no honest structural check can.

It is not an engine, and does not ask one

Nothing here queries ChatGPT, Gemini or anything else, and no result is a prediction of what one would say. That is deliberate: this check costs no engine calls, which is why it has no run limit. What an assistant does with a well-shaped page is not something we can see or promise.

The thresholds are ours

Forty per cent question headings, forty-five words before a sentence is called long, one hundred and twenty before a paragraph is called a wall. These are editorial conventions we picked and printed. No AI company publishes numbers like these, and anyone quoting some is inventing them.

It reads the HTML, so unusual markup reads badly

A page that builds its content with scripts, or lays it out in tables with no headings and no paragraphs, gives this very little to read. It says so and refuses to name a verdict rather than scoring the page down for being built unusually.

Fair questions

Short answers.

Does this tell me whether ChatGPT will quote my page?

No, and nothing on this page claims to. This check never contacts an AI engine. It reads how your page is built — where the questions are, what follows them, whether a paragraph stands up on its own. A page an assistant can lift an answer from is a page you control; whether it does is not.

What does “answer-shaped” actually mean?

That an answer can be lifted out of the page in one piece: a heading that asks something, the answer directly under it, and paragraphs that still make sense once separated from the ones around them. The six checks and their weights are printed on this page, and the total is their sum with nothing added.

Why does the check run on your server instead of in my browser?

Because a browser is not allowed to read a page from another website. That is a rule browsers enforce, not a choice we made. The address you type is sent to our server, which fetches the page — we say so rather than claiming a privacy this tool cannot have.

Can a page score well and still be useless?

Yes, and we tested exactly that. A page whose every answer restated its own question scored 97% of its measurable points. Structure is all this reads, so treat a high result as one obstacle removed, not as evidence the page is worth quoting.

My page has no question headings. Is that a fail?

It is a measured zero on the first check and it makes the second check unrun, because there is no question to test an answer against. Plenty of good pages are written that way. It means an assistant matching a question to a heading has nothing to match, which is the thing you would be changing if you changed anything.

These five answers are also the ones a machine reads — written once, used in both places.

Related tools

The rest of the question.

Render Check

Shape is moot if the content is not there before the scripts run. See what a plain fetch receives.

runs on our server

FAQ Schema Generator

Turn the questions on your page into the markup that says, in machine terms, that they are questions.

runs in your browser

AI Readiness Score

The technical checks bundled into one score you can take apart, with the weights printed.

runs on our server

Shape is one obstacle.
See the rest of them.

Score the six technical checks AI engines depend on — crawler access, rendering, schema and structure — in one run.

Check my site