The method, published
How AI visibility is earned.
The Answari methodology is a six-phase sequence for making a business visible in AI answers: baseline measurement, crawler access, entity trust, on-page optimisation, authority building, and continuous reporting. It is published here for scrutiny, and this site applies it. Publishing and applying a method are starting points; repeatable results are the evidence needed to judge it.
First principles
What are AI engines actually looking for?
When someone asks ChatGPT, Gemini, Claude or Perplexity for a recommendation, the engine builds its answer from sources it can understand, trust and cite. Two levers sit underneath everything we do:
Lever one
Machine-readability
Make the site easier to interpret with appropriate structured data, clean headings, answer-first copy and useful HTML tables. These improve clarity; their effect on citations must be tested rather than assumed.
Lever two
Entity trust
Check that the business presents accurate, consistent facts across its website and relevant public profiles, and that readers can follow supporting sources. This describes the information we can verify; it does not establish what a product privately knows or how confidently it will recommend the business.
Every tactic in the six phases below serves one or both levers. Anything that serves neither is decoration, and we do not sell decoration.
The six phases
The order matters.
Each phase has a clear “done when”. The sequence guides the investigation, while related work can progress together.
Baseline and measurement - always first
We agree the questions, direct-test products, settings and repeat schedule before measurement. Completed answers are assessed for brand mentions, source links, prominence and factual accuracy. Missing or blocked runs stay separate from negative findings. Branded and non-branded results are reported only for the groups tested. Platform analytics and attributable enquiries form separate evidence, and public profile checks do not prove an engine’s internal entity recognition.
the dated baseline includes its captured evidence, test coverage, denominators and unresolved gaps.
Crawler access and technical foundation
Crawl access is one technical foundation, not proof of inclusion in an answer. Search access and training controls have different roles: for example, OAI-SearchBot supports ChatGPT search, while GPTBot controls training access. We check relevant robots.txt rules and CDN or firewall responses separately, alongside page speed and indexation. A request using a crawler name is a diagnostic check; verified crawler traffic or the engine’s own tools provide stronger access evidence. A blocked crawler does not establish that the business is absent from an entire engine.
access, performance and indexation findings have evidence, with unverified checks clearly marked.
Entity and trust
We check material business facts across the website and relevant public profiles, and document discrepancies. Appropriate Organization and Person structured data can describe those same facts and relationships, including sameAs links where they identify the same entity. We correct what is within the agreed scope and record any third-party changes awaiting approval. No markup establishes that every engine recognises the business.
agreed public facts and markup are checked; remaining discrepancies and external dependencies are recorded.
On-page optimisation
Priority pages should answer the reader’s main question clearly, support specific facts and use suitable headings, comparisons and accessible text. A short opening answer can be useful; there is no required 40-60 word formula. We use question-and-answer sections where they help readers, with any structured data checked against the visible content. Dates reflect real reviews or material updates. Citation outcomes are tested separately.
agreed pages answer their intended questions, support material claims and pass the specified checks.
Authority and citations - the slow compounder
We identify relevant public profiles, independent sources and opportunities to publish useful original evidence. We check accuracy and record which sources appear in the captured answers. Genuine editorial coverage is earned through work others find useful; it cannot be guaranteed or substituted with purchased endorsements. Subsequent mentions and citations are measured without assuming any one action caused them.
the agreed source and profile review is complete, with actions and observed outcomes reported separately.
Measure, report, iterate - continuous
We repeat the agreed test, compare like-for-like results with the baseline and record any changes in product conditions or coverage. Reports separate work completed, brand mentions, source citations, identifiable assistant visits and attributable enquiries. Positive, negative and inconclusive results are all retained. The next steps follow the evidence and the agreed scope.
each agreed reporting period closes with evidence, limitations and prioritised next steps.
A website citation and a business recommendation need separate evidence.
- record the answer and inspect its sources
A key distinction
Direct answers and recommendations need separate evidence.
A link to a useful page and a recommendation of its publisher are different outcomes. We assess them separately:
Direct answers
“What does treatment X involve?”
Supported by clear, useful content: answer-first writing and appropriate structured data can help explain a page. Whether an engine actually links to it as a source must be observed; schema does not guarantee a citation.
Recommendations
“Who is the best provider of X near me?”
Supported by credible evidence about the business: independent coverage and verifiable expertise can help users assess a provider. We record whether the engine recommends it and which sources support that answer.
Visibility is won question by question, not business by business. The same practice can top one answer and be absent from the next - which is what the audit uncovers.
How the numbers are made
The test conditions.
We agree a question panel and record the conditions of each test. Reports show assessed answers, missing observations and separate mention and citation denominators. Product settings, location or personalisation that cannot be controlled are stated as limitations. Platform reports remain separate from these direct observations.
See the method applied in Answari’s own AI visibility study, with recorded observations, unresolved checks and accuracy issues kept visible.
- A fixed prompt set - at least ten real buyer questions, agreed in writing before the baseline and kept unchanged between reporting periods.
- A fixed direct-test engine set - ChatGPT, Gemini, Claude and Perplexity, tested through clearly recorded product and search settings.
- Recorded session conditions - use a fresh conversation and record the account state, memory and connected-context settings that can actually be checked. State what remains unknown; a new conversation alone does not prove a clean or unpersonalised session.
- Recorded conditions - each report identifies the test date, product, available model and search settings, and any known location information. Unavailable details are marked unknown.
- Repeat runs for priority prompts - the five most commercially important prompts are tested three times during each reporting period. Inconsistent results are marked as volatile rather than hidden inside one average.
- Evidence retained - answer text, source references and scoring records are retained with the engagement records so reported figures can be traced back to the underlying run.
Google Search Console and Bing Webmaster Tools data may be used as supplementary evidence where relevant. Those figures are labelled separately because platform reporting and direct prompt testing measure different things.
On the method
Questions worth asking us.
Why does the order of the phases matter?
The sequence helps us diagnose before changing things: establish a baseline, check access, clarify the business identity, improve content, and build independent corroboration. These are related workstreams, not universal prerequisites for appearing in an answer. Measurement continues throughout.
Why does every engagement start with measurement?
Because improvement cannot be proven without a starting line. The baseline records whether you appear in AI answers today, in what position, and whether the facts are accurate - and every monthly report afterwards is judged against it. It is also, candidly, the honest way to sell: you see exactly where you stand before committing to anything ongoing.
Is schema enough to win AI recommendations?
No. Structured data can help clarify page content, but it does not guarantee citations or recommendations. Google says no special schema is required for its AI features. Useful content and credible independent sources are also relevant; we measure whether changes are followed by repeatable visibility gains rather than assume a particular tactic caused them.
Why is the retainer genuine rather than a subscription for its own sake?
AI platforms change constantly and competitors respond. A citation position is not won once; it is defended. The monthly work re-runs the visibility test, reports movement against baseline, and adapts the plan. That is ongoing work by nature, not by pricing design - and if it ever stopped producing movement, the reporting would show it plainly.
Does this replace SEO?
It sits alongside SEO. The two disciplines share important foundations - technical hygiene, useful content, crawlability and clear information.
GEO then examines what happens inside AI-generated answers: whether the business is named, which sources support the answer, whether the facts are accurate and how prominently the business appears.
Neither replaces the other. Our GEO versus SEO guide explains where their methods overlap and where the measurement differs.
Phase one, applied to you
The method starts with your baseline.
The Visibility Audit tests the agreed questions across four AI products, records business mentions and website citations separately, and reviews the other providers named in those answers. You receive the captured evidence, its limits and the next priorities.
See the servicesSee the method applied
Our review of eight conveyancing fee pages records public-page observations and their limits. For a law firm’s own captured AI answers and improvement plan, see AI visibility for solicitors.
Last reviewed: 5 September 2026