How it works

How the audit works, check by check

This page shows everything behind a Full Search Audit report: the steps from order to delivery, how we collect answers from AI assistants, every group of checks we run on your site, and how to read what you receive. Real findings from a past audit are shown anonymised. Anything marked “Illustrative example” is a made-up example of the format.

80+ AI and SEO checks
4 AI assistants, 3 runs per question
Read-only, public data for the AI audits
Written report in 3, 5 or 7 to 10 business days

From order to report in five steps

Everything happens in writing, so you always have a record. There are no calls to book.

1

Order by email

Choose your audit and order by email: the order button opens an email with 5 short questions. We confirm the scope in writing and email a pro forma invoice to your business. Your order becomes binding when you pay it; until then nothing is due.

2

Tell us about your business

A short questionnaire: your website, your brand name and any spelling variants, your markets and languages, the questions your buyers ask before they buy, up to 3 competitors and your key pages (or we propose them). For the Full Search Audit, you also send us exports from Google Ads and GA4 (we tell you which; aggregated reports, no personal data) and tell us which conversions matter to your business.

3

We agree the questions

We draft the buyer questions from your site and your answers: 1 question for the Readiness Audit, 20 for the Visibility and Full Search audits. You approve or edit them before anything runs. The delivery clock starts when the questions are approved (and, for the Full Search Audit, your exports arrive).

4

We run the answers and the checks

Each question goes to every assistant 3 times. For the Visibility and Full Search audits, we also capture Google AI Overviews for the same searches. In parallel we run the site checks on your key pages and, for the Full Search Audit, review Google Ads and GA4. Every piece of evidence is saved with its date.

5

You get the report

A written PDF report in one language of your choice (English, German, French or Polish), with a prioritised plan your team can act on.

AuditDelivery
AI Search Readiness Audit3 business days
AI Search Visibility Audit5 business days
Full Search Audit7 to 10 business days

Business days, counted from when you approve the questions (and, for the Full Search Audit, send your exports).

What each audit runs

Readiness$149Visibility$290Full Search Auditfrom $490
Buyer questions12020
AssistantsChatGPT, Perplexity, Claude, Groksamesame
Runs per question on each assistant333
AI answers collected12240240
Google AI Overviews for the same searchesnoyesyes
Competitors comparednoup to 3up to 3
Site checks (technical SEO, crawlers, structured data, facts, trust, citability)yes, public reviewyes, 5 key pages in depthyes, 5 key pages in depth
Ready-to-paste JSON-LDno3 biggest gaps3 biggest gaps
Rewritten passages on your own pagesno3 examples3 examples
Google Ads and GA4, from your exportsnonoyes
Report languageone of your choice: English, German, French or Polishsamesame
Outputprioritised fix list, free fixes markedreport with a 30-day plan, grouped by type of fixreport with one combined 30-day plan, grouped by type of fix

Prices exclude VAT; VAT is added where applicable. EUR prices: €129, €249, €429.

How we collect AI answers

We ask each assistant the questions your buyers are likely to type, and record exactly what came back.

Where the answers come from

Answers are collected through the official developer APIs of OpenAI (the models behind ChatGPT), Perplexity, Anthropic (Claude) and xAI (Grok). Each request uses the assistant’s own built-in web search, so the answer is based on live web results and includes the sources the assistant cited.

Three runs per question

AI answers change from one run to the next, so every question is asked 3 separate times on each assistant. We report how often your brand and each competitor appeared, for example “mentioned in 2 of 3 runs”, rather than a single snapshot. We do not lower the randomness settings, because real users do not either.

What we record for every answer

  • The assistant and the exact model version the API reported
  • Date and time of the run
  • The country and language we requested. Questions are written in the language of your market, and where the provider supports it, the search location is set to your target country
  • Whether the assistant ran a web search, or answered from its own memory
  • Every cited URL, and whether any of them is yours
  • Every brand mentioned, including your competitors
  • Any statement about your company that does not match your own website

Google AI Overviews

AI Overviews are part of Google’s results page, not a chatbot, so we capture them separately: for the same questions, we collect live Google result pages through a search data provider and record whether an AI Overview appears, whether it cites your site and which other sites it cites, with the date and country of the search. Google does not show an AI Overview for every search, and availability differs by country and language. When none appears, the report says so.

Why an API answer is not the same as the app

An API answer is close to, but not the same as, what a signed-in person sees in chatgpt.com, perplexity.ai, claude.ai or Grok. The consumer apps add personalisation, chat history and their own instructions, and they may use a different model version, different search settings or the user’s own location. The API is the closest reproduction that can be collected in a consistent, dated way. That is why every sample records its settings, and why we describe results as observations at a point in time.

Which assistants we do not sample

Google Gemini and Microsoft Copilot are covered in the platform readiness review below, but we do not sample their answers.

Example: one run, as it appears in the report

Illustrative example
Question
Q07 of 20 "Which payroll software should a 200-person company in Germany shortlist?"
Assistant
ChatGPT (API, built-in web search on)
Model
as reported by the API for this run
Run
2 of 3
Date
23 Sep 2026, 10:42 UTC
Settings
country DE (requested), language English
Web search
yes, 2 searches
Brands named
Acme Payroll, Globex HR, Initech People
Your brand
not mentioned
Cited URLs
acme-payroll.example/germany, globex.example/pricing, hr-reviews.example/payroll-de
Wrong facts
none (your brand was not named)
Across runs
mentioned in 0 of 3 runs on ChatGPT

Illustrative example. Invented question, invented companies.

80+ checks, grouped by what they protect

We count a check only when it gets its own pass, fail or score in our method, and we count each check once even when two areas use it. The 80+ counted checks cover technical SEO, content that AI can quote, trust and content signals, platform readiness, entity presence, llms.txt and fact consistency. AI crawler access, structured data and the AI answer sample come on top of that count.

AreaChecks counted
Technical SEO37
Trust and content signals29
Platform readiness7
Fact consistency7
Citability5
llms.txt3
Entity presence (Wikipedia, Wikidata)2
Total90

From a real audit = a finding from a real audit of a B2B infrastructure company, anonymised.

Illustrative example = invented to show the format.

AI crawler accessTechnical SEOContent AI can quoteStructured dataTrust signalsFact consistencyllms.txtPlatform readiness
Access

AI crawler access

If an AI crawler cannot fetch your pages, nothing else in this report matters. We check what each named AI crawler is allowed to do, and whether your server actually lets it in.

What we check

  • Your robots.txt rules for the named AI crawlers of each operator, listed by purpose. For example: OpenAI (GPTBot for training, OAI-SearchBot for ChatGPT search, ChatGPT-User for requests made by users), Anthropic (ClaudeBot, Claude-SearchBot, Claude-User), Perplexity (PerplexityBot, Perplexity-User), Google-Extended and Applebot-Extended
  • A live request to your key pages with each crawler’s user agent, to catch firewall or CDN rules that block them even when robots.txt allows them
  • Page-level blocks on key pages: noindex, nosnippet and X-Robots-Tag headers

What we say plainly:

Some providers verify crawlers by IP address, which a user-agent test cannot reproduce. Where a firewall is involved, we state what we could and could not confirm.

Google-Extended controls whether your content is used for Google’s Gemini models. Google states that it does not affect Google Search, including AI Overviews, so we do not treat it as an AI Overviews switch.

From a real auditCrawlersHigh

A rebuilt site was ready to launch with a robots.txt that blocked every crawler and without its llms.txt. We flagged it as a launch blocker with a go-live checklist, and the site went live with crawler access open.

SEO · 37 checks

Technical SEO

AI answers and search rankings overlap. The same basics decide whether search engines and AI systems can find, render and trust your pages.

What we check

  • Crawlability (4): robots.txt is valid, XML sitemap, click depth of key pages, accidental noindex
  • Indexability (4): canonical tags, duplicate versions (www, http, trailing slash, parameters), pagination, hreflang for language versions
  • Security (7): HTTPS and certificate, HSTS, X-Content-Type-Options, X-Frame-Options, Referrer-Policy, Content-Security-Policy, Permissions-Policy
  • URL structure (3): clean URLs, logical hierarchy, redirect chains
  • Mobile (4): viewport, no horizontal scrolling, tap target size, font size
  • Rendering without JavaScript (3): main text, meta tags and structured data, and internal links present in the raw HTML that crawlers download
  • Speed, lab test (2): Largest Contentful Paint and Cumulative Layout Shift from a Lighthouse mobile test on key pages
  • Page speed and server (7): time to first byte, page weight, images, JavaScript bundles, compression, cache headers, static files served from a CDN
  • Noted, not scored (3): Link headers (RFC 8288), Markdown versions of pages, IndexNow

What we say plainly: Speed figures are lab measurements from Lighthouse, not field data from real Chrome users. We do not track keyword rankings and we do not analyse backlinks.

From a real auditTechnicalFree fix

Every page’s social card named its author as a CMS login name rather than a person. That is a weak trust signal and a free hint for anyone guessing admin accounts, and it is one toggle in the SEO plugin.

Citability · 5 criteria

Content AI can quote

Assistants lift short passages, not whole pages. We review your key pages passage by passage and score how easy each one is to quote accurately.

The 5 criteria

  1. Answer first: the passage answers a likely question in its opening sentence
  2. Self-contained: the paragraph makes sense on its own, without the text around it
  3. Readable structure: clear headings, lists and tables where a comparison is being made
  4. Specific facts: numbers, names and dates with a named source, not vague claims
  5. Original information: data, examples or explanations that only you can give

What you get: your strongest and weakest passages named, and, in the Visibility and Full Search audits, 3 rewrite examples written on your own pages.

From a real auditCitabilityMedium

On the heaviest pages, 12 to 20 percent of the text existed two or three times in hidden mobile and tablet copies of the same sections. A text extractor reads every copy, so the same passage competes with itself.

Entity

Structured data and entity clarity

Structured data (JSON-LD) states your facts in a form machines read without guessing. We check that it exists, that it is correct, and that it says the same thing as the visible page.

What we check

  • An inventory of the JSON-LD on your key pages, read from the raw page source, with syntax errors and conflicting statements listed per page
  • The types that matter for your kind of business, for example Organization, Person, WebSite, BreadcrumbList, Article, Product, SoftwareApplication, LocalBusiness, Service and Offer
  • Whether your organisation is described the same way everywhere: name, description, logo, contact details, and links to your official profiles (sameAs)
  • Markup that says something the page does not, such as prices, ratings or offers that are not visible to people
  • Types and features that search engines have retired or restricted, flagged so you do not spend time on them
  • Your entity on Wikipedia and Wikidata, checked through their official APIs, with a manual check that any match is really you

What you get: in the Visibility and Full Search audits, ready-to-paste JSON-LD for your 3 most important gaps, filled with your real data only. We never add ratings or reviews you cannot show on the page.

What we say plainly: Google states that no special structured data is required to appear in its AI features. We use structured data to make your facts unambiguous, not as a ranking switch.

From a real auditSchemaHigh

The homepage’s structured data advertised a free trial, while the visible page, the FAQ and the pricing page all said there is no free trial. The same markup carried a price with no unit, which a machine reads as the whole service costing that amount.

Trust · 29 checks

Trust and content signals

Search engines and AI systems look for evidence that real, qualified people stand behind a page. We check the signals a reader, or a machine, can see on your key pages.

What we check

  • Experience (6): first-hand writing, your own data, case studies with results, screenshots or other proof, specific examples, showing how the work is done
  • Expertise (6): author credentials, technical depth, stated methodology, claims backed by data, correct terminology, author pages
  • Authority (4): mentions of your authors in the press, awards, talks, publications
  • Trust (8): contact details, privacy policy, terms, editorial policy, a clear business model, customer reviews or testimonials, accuracy of claims, disclosure of affiliations
  • Content quality (5): depth of coverage on your topic, publication and update dates, readability, length suited to the page type, patterns of low-quality automated text

What we say plainly: we report patterns of low-quality automated text as signals only. We never state who or what wrote a page.

Illustrative exampleTrustMedium

The five most visited guides carried no author name and no publication or update date, so neither a reader nor an assistant could tell who stands behind the advice or whether it is current.

Facts · 7 fact types

Fact consistency

AI assistants repeat whichever version of a fact they find. When your own sources disagree, you cannot control which one they pick.

What we check

7 types of key facts (prices, plans, limits, locations, founding date, team, contact details) compared across your pages, your structured data, your llms.txt if you have one, and up to 3 of your profiles elsewhere, for example LinkedIn or a review site.

What you get: every contradiction listed with the exact URLs and quotes. Some platforms restrict automated access; anything we could not verify is marked “not verified”, never guessed.

From a real auditFactsHigh

A B2B infrastructure company told AI assistants two opposite things about onboarding. Four pages and all three of its AI-facing text files said there is no self-serve sign-up, while another page said “Registration is open” and a public sign-up form was live.

llms.txt · 3 checks

llms.txt

llms.txt is a proposed text file that summarises a site for AI systems. The major AI search products have not confirmed that they read it, so we treat it as optional, never as a priority fix.

What we check, if you have one

Broken links, outdated facts, and conflicts with your website.

From a real auditFactsHigh

The company’s AI-facing text files still carried last year’s prices and a free-trial offer the website no longer made, and linked to just one of its roughly 70 pages.

Platforms · 5 platforms, 7 checks

Readiness per AI platform

Each AI search product documents its own requirements. We review your site against what each one publicly documents: Google AI Overviews, ChatGPT search, Perplexity, Google Gemini and Microsoft Copilot.

Platform-specific checks (7)

Bing Webmaster Tools verification (Copilot relies on the Bing index), a Knowledge Panel for your brand, Google Business Profile and Bing Places for local businesses, Merchant Center for online shops, presence across Google properties such as News or Maps where relevant, and images or video that support your key pages.

What we say plainly: this is a readiness assessment, not a measurement. What assistants actually say is measured separately, in the answer sample. We do not score “ranks in the top 10”, because we do not collect ranking data.

Illustrative examplePlatform readinessMedium

The homepage carried no Bing Webmaster Tools verification tag. Verification can also be done by DNS, which we cannot see from outside, so the report marked it “not confirmed” and asked the team to check, rather than calling it missing.

Google Ads and GA4: where the budget goes, and whether the data can be trusted

Included in the Full Search Audit. A manual review of the Google Ads and GA4 exports you send. We never access or change your accounts: you get findings and estimates, and your PPC specialist makes the changes.

What we check

  • Conversion trackingEvery conversion action (status, primary or secondary, counting method, attribution) and which ones your bidding actually optimises toward; proxy goals such as page views that inflate the Conversions column; conversions counted twice between the Google Ads tag and imported GA4 events; GA4 versus Google Ads counts for the same period, with each gap marked explained or unexplained; the consent status that Google Ads and GA4 report for your conversions
  • Bidding and budgetWhether each bid strategy fits the number of real conversions it gets; impression share lost to budget and to ad rank; a reallocation plan within your current total budget
  • Where the money goesSpend and results by country, device and schedule; location settings that also reach people who only show interest in your target area
  • Search terms and placementsTerms and placements that spent money with no real conversions in the agreed period; a negative keyword list checked against every converting term, so we never propose blocking a query that brings you customers
  • Traffic quality signalsThe invalid click figures Google reports, plus patterns in your own data such as click spikes with no conversions or paid visits that end almost immediately
  • Ads and landing pagesResponsive Search Ad assets, pinning, Ad Strength and keyword relevance; weekly click-through trends per ad; the landing page experience part of Quality Score and message match on your main landing pages

What we ask for

Aggregated reports exported from Google Ads and GA4, sent by email or file link. We tell you which. No invites, no logins, no personal data.

Signals, not verdicts

Each traffic quality pattern is reported with a confidence level (low, medium or high) and the evidence behind it. We cannot see click IP addresses, and Google already filters invalid clicks before charging and credits some it finds later. So we never state that your account has no bots, and we never state what share of your budget went to bots.

Measured versus estimated

Money already spent on non-converting terms and placements is a measured amount from your exports. Any effect of a change, such as possible savings, is labelled as an estimate.

What you get

A tracking status table (working, broken, not verified), a prioritised fix list that says what to change and where, a negative keyword file (CSV), a placement exclusion list, a budget reallocation table, and a plain-language summary for someone who does not work in Google Ads.

Scope

Google Ads and GA4 only. Deepest review on Search campaigns; Performance Max is not reviewed in depth. Google Tag Manager, Meta, LinkedIn and Microsoft Ads are not included. Consent checks are not a legal review of your consent banner.

Illustrative exampleTrackingHigh

A “visited pricing page” goal was set as a primary conversion next to the real demo request, so Smart Bidding optimised toward page views. Reported conversions were several times higher than the demo requests the sales team received.

How to read your report

Two numbers, never mixed

Readiness Score

62/100

ExampleOur assessment

Observed mentions

14/240 answers

ExampleWhat assistants actually said

What it is
Our assessment of your site, 0 to 100
What assistants actually said, counted
Based on
The site checks above
The AI answer sample
What it is not
A metric from Google, OpenAI or any platform
A market share or a ranking

Example values are illustrative.

The report shows how the Readiness Score is built. The AI answer sample is not part of the score, so a good score never hides a poor result in the answers, and the other way round.

Severity labels, in findings

High

Blocks or distorts what AI assistants and search engines can say about you.

Fix first
Medium

Lowers the quality or trustworthiness of what they find.

Low

A smaller issue. Worth fixing when you next work on that page.

Effort labels, in the 30-day plan

Effort: free

A setting or a small edit that takes minutes and needs no development.

Marked separately: the easiest progress you can make
Effort: low

A few hours of work, such as page titles or a block of markup.

Effort: medium

A few days, or outside help: development or new content.

Evidence with every finding

HighGPTBot receives 429 from the serverIllustrative example
Page URL
yourcompany.example/
Header, rule or markup
HTTP 429 for GPTBot, curl, 3 runs
Quote or AI answer
Not mentioned, 0 of 3 runs
Date we saw it
2026-09-23

Your team can check every finding before acting on it.

The 30-day plan, grouped by type of fix

  1. Week 1Technical fixes

    Effort: freeLet AI crawlers in

    Effort: freeAdd missing security headers

    Effort: lowShorten long page titles

  2. Week 2Entity and facts

    Effort: lowOrganization JSON-LD with all your profiles

    Effort: freeFix facts that contradict each other

    Effort: mediumNamed authors with bios

  3. Week 3Quotable pages

    Effort: mediumAnswer-first openings on key pages

    Effort: lowQuestion headings buyers actually ask

  4. Week 4Proof outside your site

    Effort: mediumReviews and listings AI already cites

    Re-audit, on requestRe-run the same questions to measure the change

Each item says who usually does it: developer content marketing

Also in the report

  • A plain-language summaryWhere you stand, the biggest blocker and the three most important actions
  • Ready-to-use materialrobots.txt lines, JSON-LD blocks and rewritten passages (Visibility and Full Search audits)
  • The full list of pages and questions checkedThe methodology and the limitations

What the audit can and cannot tell you

Answers change.Results describe what we observed on the dates shown. A later run can differ, and a change within the normal variation between runs is labelled as possible noise, not progress.
A sample is not a market share.Share of mentions is counted within your question set and against the competitors you named. It is not a share of all conversations about your category.
API is close to the app, not identical.See “Why an API answer is not the same as the app” above.
Part of the score involves judgement.Some checks are scored with the help of an AI model and reviewed by a person, so a difference of a few points between two audits can be noise.
Public data only for the AI audits.Where a platform blocks automated access, or where something cannot be seen from outside (for example DNS verification or firewall rules by IP address), we mark it as not verified.
No guarantees.Nobody controls what an assistant says or which sources it picks. We do not promise that you will appear in ChatGPT, Google AI Overviews or any other AI answer, or reach any ranking, traffic or revenue.
What we do not do:we do not track keyword rankings, we do not analyse backlinks, and we do not make changes to your site or your ad accounts. Fixes are not part of any audit.

See what AI tells your buyers about you

Start with the Readiness Audit for $149. Then get 30% off a second audit ordered within 30 days of your first report.

Get your audit
See audits and prices

Questions before you order? Email hello@fullsearchaudit.com and Vadym answers in writing.