How we build your communication with AI - and keep it current
Six steps from strategy to a repeatable before/after measurement. They follow a free AI Visibility Audit: steps one to five are the one-time GEO Launch, step six is AI Visibility Growth, which starts after Launch. Packages and prices are onservices and pricing.
The measurement itself is described in detail in how we check whether AI recommends your company.
The six steps
Strategy
Which clients you want AI to bring and which you do not, the 8–10 buyer situations that matter most, your markets and limits. Revisited when the business or the results change.
- The brief every later step is checked against.
- A named person from the team, with you. Nothing is automated here.
Audit & Baseline
Part of GEO Launch, after Strategy. We lock 25–45 distinct buyer questions, worded the way clients speak, and ask each of them in all five AI tools - ChatGPT, Claude, Gemini, Perplexity and Google AI Mode (125–225 first answers), plus up to two more answer surfaces where agreed for your market - without naming you, then about you by name; branded and unbranded results are recorded separately. Every answer is stored with engine, model, date and sources. Alongside: your site's technical accessibility, its information architecture and the external sources that speak about you. From that we name three to five causes we can act on, and the order in which Launch addresses them.
- A dated baseline - how often you are in the answer (AI Choice Visibility), who the engines choose instead, which sources they trust, what they get wrong - with the causes in plain language and the priorities for Launch. No composite score.
- People write the questions, put them to each AI system, read every answer and write the diagnosis; software checks the technical layer of the site.
Map & Architect
One Fact Bank for the company: services and the buyer situations they answer, experts tied to the work they actually did, cases and evidence, the external sources that confirm them, and how each result will be measured.
- A map of what goes where, and why, before anything is built.
- The team, with your people for facts and approvals.
Build & Connect
The website and its structured data, content, expert profiles, cases, and the external records - registries, directories, partners, media - connected so that each points to the others and says the same thing.
- Pages and records that agree with each other and can be cited.
- The team.
Launch & Verify
We check that the system can actually be read: crawlers are let in, the page says the same with and without scripts, nothing contradicts itself, old and unpublished pages are not left for AI to find, and the measurement is in place.
- A launch checklist signed by a person, with the open items named.
- Programs check; a person signs.
Measure & Grow
AI Visibility Growth, after Launch. The same locked questions, the same engines, on new dates. We compare before and after - citations and sources, who moved, what visitors did and which enquiries were qualified - choose the next priority and strengthen it. Engines change; the measurement does not.
- A before/after table each cycle, with the sources and the companies named, and one decision for the next cycle.
- Programs repeat the run; a person reads the difference and signs the report.
From AI answers to decisions
To inform strategic content decisions, we look at how leading AI systems answer the questions buyers ask. We compare those answers with each other and with primary sources, then make our own decision. AI answers show the narrative a buyer meets. They do not decide our position.
Buyer questions
We write the questions in the words buyers use, grouped by situation: choosing a provider, comparing options, checking a firm.
AI answers
We record the answers and the conditions we can observe. Missing or unverified records are flagged separately. Incorrect statements in an answer are retained and analysed as findings.
Compare and check
Where the answers agree, where they differ, and which claims hold up against primary sources. A model's explanation of how it chooses is an observation, not proof of how it ranks.
Our decision
Our position and its basis: professional experience, project data and external evidence.
Changes
What we connect or clarify on the website and in external sources: a service page, an expert profile, a case, a reference. Each material claim has a clear basis.
Re-test
The same questions on a new date. We note any change in conditions, report what changed and what did not, and do not credit a shift in answers to a single edit unless the test was designed to show it.
What AI answers tell us - and what we do with it
Before important content is written, we read how leading AI systems currently frame the subject your buyers ask about. That shows the picture a buyer meets. It is research input, not the truth, and it does not set our position.
- Selection criteria
- Objections and risks
- Expected proof
- Sources
- Follow-up questions
- Disagreements
We then compare that picture with professional experience, project data, direct observation and external evidence. The output is a short chain: what the answers say → our position → the basis it needs → the change on the website.
Example: a contested recommendation - llms.txt
In our research run dated 5 October 2026, the answers differed on the llms.txt file in priority rather than in kind: some described it as unproven or low priority, others listed it among recommended steps, in some answers as optional.
llms.txt is not a priority.
Google's documentation says no special files are needed to appear in AI Overviews and AI Mode in Google Search. That statement covers Google Search, not every AI system.
The work goes first to pages AI systems can read, consistent facts and independent sources.
Start from the trust you already have
For an established firm, the hardest asset - trust - often already exists. The audit asks two separate questions about it.
How much real trust already exists, and is it connected to the service and market you want to win?
Do the AI answers we observe appear to use that trust?
How the comparison sets priorities
Trust is strong and the answers reflect it
Trust is strong, but the answers do not reflect it
Trust is thin or tied to a different service
ChilliTax shows the second pattern in its starting point: 5.0 stars from 79 Google reviews and recognised by name in all five AI systems tested, yet not named in 25 unbranded answers about choosing a corporate-tax provider (1 September 2026). ChilliTax baseline details.
How a repeat measurement changes the plan
Every change we make starts as a working assumption: what we expect it to improve, and why. The repeat measurement tells us what to do with it next.
Assume
Test
Read the result
Continue
The signal improved: keep the change and look for ways to strengthen it.
Adjust
Replace
Three things we keep apart
- AI recommendations
- What the answers say: mentions, recommendations, the competitors named and the sources cited.
- Enquiries
- Relevant contacts and where the people who made them say they found you.
- Clients
- Signed work - the commercial result, recorded separately from what AI answers say.
Our own AI Narrative Mapping
Dated 5 October 2026: 38 research questions - 35 about GEO and choosing a provider, three about GO!GEO and its website - each asked in ChatGPT, Claude, Gemini and Perplexity. The answers show how these AI tools describe the field and our company, and we use them to decide what to make clearer. This study maps narratives; it does not measure how often AI tools recommend GO!GEO.
- Questions about GEO and choosing a provider
- 35
- Questions about GO!GEO
- 3
- AI tools
- 4
After launch
Your business changes every week. Your digital presence usually doesn't. You give us the new facts - a project, a client, an award, a talk, a result. We place each one where it strengthens the system:
- service, programme, sector and country pages
- expert profiles and the real roles of people
- cases, results and evidence
- links to independent sources
- channel-ready posts for company and employee LinkedIn
- a repeat AI measurement and the next priority
What the GEO Launch audit looks at
Three questions a client would ask in plain words, and what we measure behind each. The technical layer is listed so you can hand it to your developer; the report itself is written for you.
Can the engines read you?
AI crawlers are allowed in
robots.txt, sitemap and status codes decide whether ChatGPT, Gemini or Google's AI ever see the page.
The page says the same thing with and without JavaScript
Many engines read the raw HTML; what appears only after scripts run may not exist for them.
Structured data matches the visible text
Organization, Person and Service records tell the engine who you are - if they contradict the page, neither is trusted.
Do the facts about you agree?
Name, address and roles are identical everywhere
Register, site, LinkedIn and directories are compared; one differing spelling is one more reason to pick someone else.
Independent sources say what you say
Most brand mentions in AI answers come from third-party pages, not from the brand's own site.
Old and unpublished pages are out of reach
A page you consider outdated or not yet live may already be what the engine finds and repeats about you.
Are you in the answer?
Unprompted buyer questions, repeated across engines
Two or three runs per question per engine - a single answer is an anecdote, a frequency is a measurement.
The brand question, asked second
Shows whether the engine knows you but does not choose you - a different problem with a different fix.
Who is named instead, and from which sources
The competitors' evidence is the map of what you are missing.
What we promise - and what we do not
We promise
- A locked question set that does not change after an inconvenient result.
- Every answer stored with engine, model, date and sources - the run can be repeated and audited.
- Causes we can act on, or a clear statement that we cannot see one.
- A named person who ran the check and signs the report.
We do not promise
- A specific answer from a specific AI. Nobody honest can promise it.
- A single score. The metrics are reported separately, each with its definition.
- Results in two weeks. Citation typically moves in 30–90 days, and we will not pretend otherwise.
- Volume. We publish what can be checked, not thirty articles a day.
- A secret scheme. No llms.txt file or single piece of markup makes AI recommend anyone.
- Your claims as proven facts. What you say about yourself stays labelled as yours until a source confirms it.
- A site built only for machines. People still read, compare and decide.
Not for you if
- You need a result in two weeks.
- You want thirty articles a day rather than pages that can be verified.
- Your site is a placeholder and the business is not described anywhere yet - the facts come first, visibility second.
- You need a guarantee that ChatGPT will name you.
Start with a free AI Visibility Audit
A first look at one priority service: how leading AI assistants understand and recommend your business, and which competitors they name instead. The full baseline is part of GEO Launch.