Paste one URL and see how the page reads to Google and to AI assistants: its SEO basics, whether its structure is easy to quote, and whether it shows who wrote it. This is not an AI-text detector, and it never guesses who or what wrote the words. It runs up to 42 of the page checks Verand's site audit runs on customer sites and lists every finding here.
27 page checks on any page, 42 on an article. One GET of the URL as VerandBot/1.0, redirects followed, the served HTML only: no JavaScript is run and no paid render is bought. The 15 article checks (answer structure, author, dates, citations) run only when the page reads as an article: Article schema, a /blog/-style address, or enough article signals. A homepage never counts.
A pass earns full credit, a miss half, and a critical miss (no title, a noindex tag) none. Each dial is credit over checks, times 100, and the overall blends them 45 / 20 / 35. Same formula as Verand's audit, over this page's checks only; site-wide checks are left out of both the list and the score.
Anything a script adds after load. Anything past the first 100,000 characters of HTML. Site-wide checks (duplicate titles, orphan pages, link depth) need a crawl, and robots.txt and llms.txt are site files with their own tools. It measures structure, not whether an assistant will cite the page: no check can promise that.
One fetch, one outline, up to 42 questions. The card is the tool in motion on an example article, looped, and each step lights up while the card is doing it.
The tool fetches the URL you paste as VerandBot, following redirects, and keeps what the server sent. No browser runs the page, so text a script injects later is not in what gets checked. That is also roughly what a crawler that skips JavaScript receives.
The page is parsed into an outline: title, meta description, headings, links, images, JSON-LD, and the main content with the nav and footer stripped away. Then one decision: does it read as an article? Only an article gets the 15 checks for answer structure, authorship and sources.
Each check asks one fixed question with a fixed threshold: is the title under about 60 characters, is there exactly one h1, does the first paragraph run 30 to 120 words, is there a named author in the schema. A pass or a miss, never a model's opinion.
Passes and misses become the SEO, AI answers and E-E-A-T dials, then an overall. Every miss is listed with what it measured, and each group links to the single-check tool that goes deeper on it.
Readiness is a question about the page, not about the writer. Here is what a page needs before Google or an AI assistant can use it well, which of those things this checker reads, and where the honest limits sit.
Search "AI content checker" and most of what comes back are detectors: tools that estimate whether a model wrote a piece of text. This is a different question. A readiness check asks whether a page, whoever or whatever drafted it, is built so that a search engine and an AI assistant can fetch it, understand what it is about, lift a clean answer from it and tell who stands behind it. A page written entirely by hand can fail every one of those. A page drafted with a model and edited by an expert can pass them all.
That makes the checks concrete. There is a title and a description that fit where they are shown. There is one main heading and a sensible run of section headings. The text is in the HTML the server sends. The first paragraph answers something. Key terms are defined. The author is named in a way a machine can read, with dates beside the content. Each of those is a thing you can look at and fix, which is the only kind of AI readiness checker worth running.
Readiness starts before the content. A page an assistant's crawler is turned away from, or one that tells search engines not to index it, is out of the running however good it is. This checker reads the page-level signals: whether it is set to noindex with a robots meta tag, whether it redirects, whether it has a canonical tag naming the preferred URL, and whether it is served over HTTPS. The site-level question, which AI crawlers your robots.txt lets in by name, is a separate file with its own tool: the AI Crawler Checker reads it for the whole site, and the AI Crawler Access Checker answers it for one URL. Run one of those first if you have never looked.
This is the table-stakes point most people skip. Many sites build the visible page in the browser: the server sends a shell and a script fills in the article. A person sees the full page. A crawler that does not run the script sees the shell. Googlebot renders JavaScript, on its own schedule. Most AI crawler operators do not document whether theirs does, so you cannot count on it. The safe position is that the answer you want quoted should arrive in the first response.
That is exactly what this checker reads. It makes one request, keeps the HTML the server returned, and runs every check against it with no browser involved. If your page scores oddly low on word count or headings while looking complete on screen, that gap is the finding: the content is there for people and missing for anything that reads the raw response.
Headings are the outline a machine sees first. The checker asks for exactly one h1, flags a page with none and a page with several, and notes when the h1 repeats the title tag word for word. Among the article checks, the structure check asks for at least two h2 sections. Section headings that name the question they answer give an assistant a clean place to cut, which is why the checker also looks for a page of question-shaped headings with no FAQPage markup. The Header Tag Checker shows the full heading tree if you want to see the outline itself.
The AI answers dial is made of six checks, and it only runs on a page that reads as an article. It measures one thing: structure that is easier to quote. It does not measure, and cannot measure, whether any assistant will actually cite the page. Nobody outside those companies can see that decision, and a tool that claims to score it is guessing. What can be counted is whether the page offers the kinds of passages that are simple to lift intact. These are the six, with the thresholds they use.
| Check | Passes when |
|---|---|
| Quotable sentences | At least three paragraph sentences of 8 to 40 words carry a number or a plain claim verb (is, means, requires, reduces). Standalone statements survive being lifted out of context. |
| Answer-first structure | The first paragraph runs 30 to 120 words, there are two or more h2 sections, and there is at least one list or table. All three, or it is a miss that names which part failed. |
| Defined terms | At least two sentences define something ("X is a", "X refers to", "X is defined as") or the page uses dfn or dl markup. |
| Takeaways block | A heading such as Key takeaways, Summary or TL;DR is followed by a list of three or more items. |
| Article schema | The page declares Article, BlogPosting or NewsArticle in JSON-LD, so its headline, author and dates are stated, not inferred. |
| FAQ markup | A page with three or more question-shaped headings also carries FAQPage schema. |
The 30-to-120-word intro is Verand's rule of thumb, not a published standard. You will see other tools say the answer belongs in the first 100 words; that figure is a heuristic too, and nobody has shown a cut-off. The point both are reaching for is simple: the first paragraph should answer the question the page is named for, instead of warming up to it.
JSON-LD is a block of machine-readable facts in the page's source: this is an article, this person wrote it, it was published on this date, it was changed on that one. The checker confirms every JSON-LD block parses, because one malformed block means none of its markup can be read, and on articles it asks for Article schema, author entity links (sameAs) and citation markup. What structured data does is make the page easier to read correctly. It is not a lever that puts a page into AI answers, and it cannot state anything the visible page does not. The Schema Checker lists every type a page declares.
On an article, ten checks feed the E-E-A-T dial: a named author in the schema, the author's credentials, first-person experience in the prose, a reachable author bio, published and modified dates, citation schema, whether the article cites sources at all, whether any of those sources are authoritative, entity links on the author, and a copyright or "last updated" year that is not stale. These are the signals a reader, a search engine and an assistant all use to decide whether the page is someone's considered view or anonymous filler. The E-E-A-T Checker runs the same set on its own, with more on what each one reads.
Most readiness advice pushes one way: shorter, more direct, answer first. For a financial adviser, a law firm or a clinic, that pressure has a cost. The sentence that makes an answer extractable is often the sentence that has lost its qualifier: "a Roth conversion avoids tax" instead of "a Roth conversion can reduce tax later, depending on your bracket now and in retirement." The first reads as more quotable. The second is the one your regulator expects.
Two parts of that this checker does read. Dates: the freshness check wants both datePublished and dateModified in the Article schema, and a separate check flags a copyright or "last updated" year two or more years old, so the answer does not travel without its age. What it does not read is whether a required qualifier sits next to the answer, or whether the disclosures your industry expects are on the page at all. Those need a ruleset for your industry, which is what the Disclaimer Checker and the Marketing Compliance Checker apply. Keep the answer first and keep its condition in the same sentence; a quotable answer that drops the condition is not an improvement.
An llms.txt file is a plain list, at the root of your site, of the pages you consider your best and most carefully reviewed. It is a site file, not a page, so this checker does not read it. It is also optional: no major assistant has confirmed it reads the file, so treat it as a tidy index you control rather than a switch. The llms.txt Generator writes one from a crawl of up to 25 of your pages.
h1. One names the page; the rest belong one level down.A high score means the page is fetchable, correctly described, cleanly structured and clearly attributed, as far as deterministic checks can see. It does not mean an assistant will quote it, and it does not mean the content is right. Accuracy, depth and whether the page answers the question better than the pages around it are editorial judgements. This on-page SEO checker clears the ground so those judgements are the only thing left to work on.
Six things that are true of this tool, each one backed by a line in the code that runs it.
The request carries a URL and nothing else. There is no account, no session and no database behind the tool, so there is nothing for us to keep about you.
Every miss comes back with what it measured, and every group names the checks that passed. Nothing is held back for a report, a call or an upgrade.
Each check is a fixed rule with a fixed threshold. No model reads the page and no score is guessed, so the same HTML gives the same result every time.
These are the page checks Verand's deep crawl runs on every page of every customer site every two weeks, called on one URL. Not a lighter demo version.
Script-built content, the HTML past the read limit, site-wide checks and citation itself are out of scope, and the card says so beside the result rather than in a footnote.
Each run is one page fetch and a set of rules, so it costs nothing and is never metered. The one limit is a courtesy to the sites being fetched: 20 checks a minute per visitor.
What readiness means, what a pass promises, and how to use the result.
Whether a page is built so that search engines and AI assistants can fetch it, understand it, take a clean answer from it and see who stands behind it. It is a property of the page, not of who wrote it: this checker does not detect AI-written text. It reads the served HTML and runs up to 42 fixed checks across SEO basics, answer structure and authorship.
The 27 checks that run on every page are the familiar ones: title and meta length, headings, canonical, noindex, HTTPS, images, links and social tags. On an article it adds 15 more: quotable sentences, an answer-first intro with sections and a list, defined terms, a takeaways list, Article and FAQ schema, and the authorship signals (named author, credentials, first-hand experience, bio, dates, citations and entity links). Those are scored as their own dials rather than folded into one SEO number.
No, and no tool can promise that. Which pages an assistant quotes is decided inside systems nobody outside can inspect, and it depends on the question, the competing pages and more. What the checks measure is whether your page removes the obstacles they can see: it can be fetched, it is described correctly, its structure is easy to quote and its author is identifiable. That is worth doing, and it is all this checker claims.
Each miss on the card says what was measured, for example the title's length or which part of the answer structure failed. Most fixes are an edit in your CMS: shorten a title, demote an extra h1 to h2, add a takeaways list, fill in the author and dates in your SEO plugin's schema settings. Each group links to the single-check tool for it, which explains that one check in depth. Run the page again after the edit; the result is deterministic, so a changed score means a changed page.
Because the answer-structure and authorship checks belong to a page, and one page is what you edit. A whole-site geo audit needs a crawl, and some checks only exist across pages: duplicate titles, orphan pages that nothing links to, how deep a page sits. Those are left off this card rather than shown as passes. Verand's product runs the same page checks on every page of a customer's site, plus the site-wide ones, every two weeks.
After any edit to the page, and after changes to your theme, page builder or SEO plugin, since those rewrite headings, schema and meta tags across every page at once. It is free with no daily allowance; the only limit is 20 checks a minute per visitor, a courtesy to the sites being fetched.
Verand writes articles from your own expertise and credentials, so there is something of yours for Google and the AI assistants to find. It then tracks where you rank and where ChatGPT, Gemini, Google AI Overviews, Google AI Mode, Perplexity and Claude name you, and gates every draft so a claim your regulator would not allow never publishes.
Content built to rank in
Google and get cited by
ChatGPT
Perplexity
Gemini
Claude, with every claim checked before it goes live.