AI answer-readiness checker

What the checker measures, and why those signals

Most AI answer-readiness checkers make you guess at a grade. This one scores a handful of things answer engines demonstrably rely on when they read your page and decide whether to quote it. First, the page has to be reachable at all - if a bot wall or cookie banner hides the content, nothing else matters. Then it looks at your heading structure: exactly one H1 stating what the page is about, and enough H2 sections that someone (or some model) can reconstruct your argument from the outline alone. Question-formatted headings get extra weight, because buyers type questions and engines resolve them against pages that visibly frame the same questions.

The remaining signals are about the machine layer. FAQPage schema should be present and parseable, llms.txt earns a few optional points, and robots.txt is checked per-crawler: GPTBot from OpenAI, ClaudeBot from Anthropic, and PerplexityBot, whose index sits behind the fastest citation-generating engine of the three. One Disallow line aimed at the wrong group has taken whole brands out of ChatGPT and Perplexity answers for months at a time - the checker names the exact bot that is shut out.

The heading rules that actually help you get quoted

An AI engine rarely reads your page top to bottom the way a person would. It extracts: the H1 for the topic, headings for the structure, and - where a heading reads like a question - the first one or two sentences beneath it as the candidate answer. That is why the checker is strict about one H1 and generous with H2s. If your page has three competing H1s, each engine sees three different possible topics. If it has one weak H1 and eight meaningless section titles, the intent of the page never lands.

Practical pattern: state the topic in the H1, make your main sections the questions a buyer would actually type, and answer each in its first sentence. A “What is...” heading followed by two paragraphs of preamble tends to lose the citation to a competitor that states the definition first. Question-formatted overlaps with how Perplexity and ChatGPT segment sources - they lift content that already reads like an answer.

FAQ schema in 2026: honest version

FAQ rich results were retired by Google in May 2026, so no, FAQPage schema will not earn you the expandable accordion result in search it once could. It is still a decent cheap win: engines that build answers from pages (Perplexity among them) can lift the question and its first answer directly from your markup, and the exercise of writing the FAQ usually produces the Q&A that should live in your headings anyway. Use the FAQ schema generator if writing the JSON-LD by hand is friction. The content matters more than the markup - but the markup is five minutes.

What to do with your score

Take the pass / improve / missing list top-down. Anything failing is usually a small edit: add an H1, rename three headings into questions, publish a llms.txt, or remove a Disallow rule. Re-run the check after each change to confirm the score moves, and then check the other pages buyers are likely to land on - a 404-level readiness on a product page is worth ten blog fixes. Once your pages are structurally sane, the next step is not more scoring: it is watching what the engines actually say about you against the competitors they name instead. Sonarvue does that across ChatGPT, Perplexity and Google on every plan, so this checker is deliberately the free door to that habit.

Fix what it names, then keep watching

This checker grades one page, once, for free. When you want the same evidence across every page that matters - and, more importantly, a record of what ChatGPT, Perplexity and Googlesay about your brand against the competitors they name instead - the Sonarvue trial scan is the next step. Same scoring philosophy underneath, plus the tracking that turns a score into visibility you can defend.