Which six surfaces does the audit cover?
The audit covers the classic search results page, Google AI Overviews and AI Mode, chat assistants such as ChatGPT, Claude, Gemini and Perplexity, voice assistants, video search, and a large language model reading the site directly through clean markup and an index file. A buyer with one question can arrive on any of them, often several in one afternoon, and each rewards something slightly different in its selection.
The reason ftes.ai audits them together is that the differences are smaller than they look. Five of the six reward a passage that answers a question completely on its own; the sixth rewards a page that loads fast and is structured clearly.
The classic search check: is the page eligible at all?
Read the page against the query a person would type. Does the title carry that query near the front, is there exactly one first-level heading, do the second-level headings progress without skipping a level, does the page link to the three most relevant neighbours, and does it load without a layout shift? These are old checks and they still decide whether a page is eligible at all, because the other five surfaces mostly draw from an index that this one populates.
The failure here is usually structural rather than editorial: two first-level headings, a heading level skipped, or a page whose internal links all point upward to hubs and never sideways to siblings.
The AI answer check: do the first forty words answer?
Read the first forty to sixty words alone, as a stranger would. Do they answer the question, name the entity, and qualify the answer once? Then read each section's opening sentence alone and ask the same. An AI answer is assembled from fragments, so the test is fragment-level and unforgiving: a passage that begins with context, history, or a hook has spent its most valuable sentences on something the assembling model discards.
The fix is the ftes.ai answer-first rule, applied to the page and then again to every section. It is the single highest-yield edit on this list, and it usually shortens the page.
The chat assistant check: does the section survive alone?
Chat assistants retrieve passages and cite sources, so the check is whether each section would survive being read with nothing around it. Take any section out of the page, read it cold, and ask whether its subject is named, its claim is complete, and its specifics are checkable. Sections that fail usually fail for one of three reasons: a pronoun where the entity should be, a cross-reference to another part of the page, or a claim with no mechanism behind it.
This check overlaps heavily with the AI answer check, which is the point. Fixing a page for one surface fixes it for the other.
The voice check: can any one answer be spoken?
Read the frequently asked questions aloud. At least two answers should be short enough to be a spoken reply — around thirty words — with no parentheses, no slashes, and no abbreviations that only work on a page. Read the questions aloud too: they should be phrased the way a person speaks rather than the way a person types. A voice assistant returns one answer, so the test is whether any single answer on the page could be that one.
The ftes.ai content verifier enforces the two-short-answers rule mechanically because it is easy to lose in editing and easy to restore.
The video and direct-read checks: chapters, and markup a crawler can read
For video, read the second-level headings as a list and ask whether they would work as chapters: five to nine of them, each independently meaningful, in an order that makes sense without the prose. A page that passes this is also a page whose sections are self-contained, which is why the check is worth running even when no video exists. Structured data for video ships only when a real video exists, and never as a way to look like it does.
For the direct read, check that the page is reachable without JavaScript, that its structured data is server-rendered inline, and that it appears in the site's index file for language models. A page whose schema is injected by a client effect is invisible to a crawler that runs no script.
What the audit is for, and what it is not
The audit is a pre-publish and quarterly check that finds structural faults cheaply. It does not predict traffic, and it does not score a page against competitors, because both would require data a publisher does not have. Its value is that it converts six vague ambitions into about fifteen minutes of reading with a specific fix behind each failure, and that the fixes do not conflict with one another.
Run it on the pages that carry commercial weight first. A library of several hundred pieces does not need every page audited every quarter; it needs the twenty that answer buying questions audited every time the product changes.