Can an answer engine reach, parse and quote your site?
Before a model can recommend you, it has to be able to fetch your pages, work out what you are, and find something specific enough to repeat. The audit checks those three things page by page and reports what it found, with the URL attached.
- Scope
- Crawler access, answerable structure, entity clarity, citable evidence, machine-readable layer, freshness.
- Granularity
- Findings attach to a URL and a category, never to a vague site grade.
- Honesty rule
- Anything with no established effect is labelled informational and left out of the scoring.
Seven things we look at, and why each one is on the list.
The audit is not a generic technical crawl with an AI label on it. Every check is here because it changes whether an answer engine can use the page as source material.
- Access
Can an AI crawler reach the page at all?
- robots.txt directives, including rules aimed at named AI and search crawler user agents
- Edge, CDN and bot-management rules that return a challenge instead of the document
- Content that only exists after client-side rendering, where the fetched HTML is effectively empty
- Redirect chains, soft 404s and status codes that differ for a crawler and a browser
- Structure
Is the page shaped like an answer?
- Headings that state the question a section answers, rather than a slogan
- A direct answer near the top of the section, not buried under three paragraphs of preamble
- Lists, comparison tables and definition blocks that survive extraction as self-contained units
- Sections short enough to be quoted without dragging in unrelated context
- Entity clarity
Is it obvious what you are and who you serve?
- A plain statement of what the product does, in the category vocabulary buyers actually use
- Consistent brand naming across pages, so one company does not read as two entities
- Explicit audience, use cases and boundaries — including what the product is not for
- Organisation details and cross-references that let an engine resolve you to one identity
- Evidence
Is there anything specific enough to quote?
- Concrete specifics — supported integrations, limits, regions, mechanics — rather than adjectives
- Documentation and changelog depth, which is often the most citable material a company owns
- Comparison and alternatives coverage that exists on your own domain, in your own words
- Claims attributed to something checkable, since unattributed superlatives are not quotable
- Technical
Is the machine-readable layer intact?
- Structured data for the entities that matter: organisation, product, FAQ, article
- Canonical tags, duplicate URLs and parameter variants that split one page into several
- Sitemap coverage and internal linking to the pages you most want read
- Metadata that matches the page body rather than a legacy template
- Freshness
Does the page look current?
- Visible and machine-readable modification dates
- Pricing, packaging and integration pages that have drifted from the product
- Dated comparisons naming competitors or versions that no longer exist
- Stale help content that contradicts the current interface
- Emerging
Emerging conventions
- llms.txt presence, location and parse validity
- Other proposed AI-directive files, reported as present or absent
- Whether a declared file is consistent with your robots.txt rather than contradicting it
Detected and reported. Not sold to you as a lever.
AnswerRank checks whether your site publishes an llms.txt file, whether it is where it should be, and whether it parses. Then it labels the finding for what it is.
That is the exact label the audit uses. We are not aware of evidence that publishing the file changes how answer engines represent a brand, and we are not going to imply otherwise to make a checklist look longer.
- The finding is informational: it does not affect any score in the product.
- If you publish one, we check it is consistent with your robots.txt rather than contradicting it.
- If the convention gains real supporting evidence, the label changes and the change will be visible in our methodology.
How we treat unproven conventions in general
AnswerRank never describes a convention as an established ranking factor, and that class of phrasing is blocked in the code that generates our measurement copy. Nobody outside a provider can verify a ranking factor for a system whose retrieval and generation steps are not observable.
So the audit separates two things: findings that block or degrade an engine’s ability to use your page, which we score and prioritise, and findings that are merely present or absent, which we report and leave unweighted.
If a recommendation cannot point at an observation in your own data, it does not get an impact rating.
Every finding names a URL.
A site-wide letter grade tells nobody what to do on Monday. The audit output is a list of pages, each with the category it failed, what specifically was wrong, and how severely it matters.
Audit findings · acme.com
| URL | Finding | Category | Severity |
|---|---|---|---|
| /pricing | Plan limits rendered only after client-side hydration | Access | Blocking |
| /product/analytics | No direct answer in the first section; three paragraphs of positioning first | Structure | High |
| / | Category described in two different vocabularies across the page | Entity clarity | High |
| /integrations | Integration list is an image; no text equivalent | Evidence | Medium |
| /llms.txt | Not present | Emerging | Informational |
| Severity | What it means | How to read it |
|---|---|---|
| Blocking | The engine cannot use the page at all — it is unreachable, empty on fetch, or serving an error. | Fix before anything else. No amount of content work outruns a blocked page. |
| High | The page is reachable but hard to extract from, or the entity it describes is ambiguous. | Associated with weaker representation in answers about this topic. Usually a bounded edit. |
| Medium | A structural or machine-readability gap that limits how cleanly the page can be quoted. | Worth batching into planned content work rather than treating as an incident. |
| Informational | An observation with no established effect, including emerging conventions we detect but do not endorse. | Reported so you know the state of it. Not weighted, not scored, not presented as a lever. |
The audit is the third question, not a standalone report.
On its own, a technical audit is a list of chores. Inside the loop it becomes an explanation for a specific visibility gap — and then something you can test.
- 01
A gap appears in tracking
You are absent from a cluster of comparison prompts on two engines, and the citation view shows a competitor-owned page taking the slot.
- 02
The audit supplies a candidate reason
Evidence suggests the relevant page is a likely contributing factor: it renders client-side, so the fetched HTML the engine reads is effectively empty.
- 03
An experiment settles it
Fix the page, freeze the prompt set, re-run the identical checks, and read the delta over a stated evaluation window.
The audit tells you what is broken and what is merely unusual. Only the experiment can tell you whether fixing it was associated with a change in how the engines answered.
See what an answer engine sees on your site.
Run a free check. No signup, no card.
No signup required for your first visibility check.
Every figure AnswerRank reports carries its sample size and how it was collected.
Read the methodology