An LLM Readability Audit for an Existing Page
Run nine checks in fix-cost order: is there a complete answer near the top, does each heading pose a question, does each section answer immediately, is every passage self-contained without pronouns reaching backwards, do tables follow one fact per cell with no bare symbols, are volatile facts dated inline, are claims specific and attributed, is the content reachable without JavaScript, and is there at least one inbound internal link from a page that is already indexed. Most published pages fail the same three — a warm-up intro, backwards-reaching pronouns, and undated volatile facts — and all three are cheap to fix without rewriting.
This is an audit for pages that already exist and are not being read by anything. It assumes the content is basically sound; if it is not, no amount of restructuring helps and the refresh-versus-new-page question comes first.
The nine checks
| Check | How to test | Typical fix cost |
|---|---|---|
| Answer near the top (editorial target, not an engine rule) | Read the opening alone — does it answer? | Low — move a paragraph |
| Headings pose questions | Read the headings as a list | Low — rewrite headings |
| Answer follows heading immediately | Check the first sentence of each section | Low — reorder |
| Passages self-contained | Search for 'it', 'this', 'they' opening a paragraph | Low — name the subject |
| Tables extractable | Read one row alone | Medium — restructure cells |
| Volatile facts dated inline | Find every number; is it dated? | Low — add as-of text |
| Claims specific and attributed | Find every 'often', 'many', 'studies show' | Medium — needs the real number |
| Content present without JavaScript | Fetch the page with JS disabled | High if it fails |
| At least one inbound internal link | Search the site for links to it | Low — add to a related section |
The three most pages fail
The warm-up intro, where two paragraphs of context precede the answer. Backwards-reaching pronouns, where a paragraph opens with 'This means that…' and the antecedent is in the previous section. And undated volatile facts, usually prices. All three are cheap, and fixing them does not require touching the argument of the page.
The check that is not about writing
Content present without JavaScript. If the copy only exists after client-side rendering, everything else on this list is moot for any retriever that does not execute scripts. It is also the most expensive to fix, which is why it is worth testing early rather than discovering after a rewrite. Our crawler retrieval guide covers verifying it properly.
The check that is not about the page
Inbound internal links. A page reachable only from a sitemap or a hub listing is discoverable in principle and deprioritised in practice. One link from an already-indexed page, in the body rather than in a footer list, is the cheapest fix on this list and the one most often skipped because it requires editing a different page.
FAQ
How long does this take per page? — Around twenty minutes to audit, and most of the fixes are minutes each. The expensive ones are the JavaScript check failing and vague claims that need a real number, because the second means going and finding the number.
Will fixing these get the page cited? — Unknown, and anyone promising otherwise is selling something. What they do is remove specific avoidable reasons a page cannot be used. That is worth doing on its own terms and is not the same as a guarantee.
Should I audit every page? — No. Audit the pages you would most like to be quoted on, which is usually a small set. Running this across a whole site is a way to spend a week on pages nobody was going to cite either way.
Put this to work on your own website.
Fenn finds what your customers ask, drafts the articles and site fixes, and measures what ChatGPT, Claude, Gemini, Perplexity and Grok say about you — with every change waiting for your approval.