Discovery
Can the right buyer find you, and are you represented correctly?
Question coverage, correct entity representation, mentions, recommendations, citations, and crawl and index eligibility.
A diagnostic. A citation is not a lead.
Methodology
How we measure demand, not just visibility
We measure three different things and never blend them into one score: whether buyers can find you, whether the right ones act, and whether your sales team accepts what arrives. Every outcome claim says how it was established.
Discovery is sampled across the major answer engines
- ChatGPT
- Claude
- Perplexity
- Gemini
- Google AI Overviews
Three layers, measured separately
Visibility is the layer most tools stop at. It is real and worth measuring, but it is a diagnostic. Rising mentions with no relevant enquiries points to a journey or offer problem, and more content will not fix it.
Buyer behaviour
Do the right people arrive and act?
Relevant visits, quote or demo starts, form completions, calls, and the discovery source a buyer declares.
Analytics show behaviour, not intent or quality.
Commercial outcomes
Did your team accept the enquiry, and did it become revenue?
Accepted opportunities, rejection reasons, opportunities won, collected revenue, and contribution where you can share it.
Needs your CRM feedback. Without it we say so.
Three attribution labels, never merged
Every claim about where an enquiry came from carries one of these labels. The weakest label that fits is the one we use.
Observed source
What was directly recorded: the form field, the referrer, the tracked call. Narrow, but it is a fact.
Reported influence
What you or your buyer say helped. Useful, honest, and not proof.
Estimated incrementality
What a suitably designed comparison or experiment supports. When the evidence cannot carry this, we do not claim it.
Where measurement sits in a mandate
Measurement is the first and last step of every programme: the baseline we agree in the mandate, and the monthly reconciliation that decides where the budget goes next.
- 01
Agree the mandate
A written brief in your terms: the objective, the buyer, the product and territory, the budget, the exclusions, and what counts as an accepted enquiry. Facts, your statements and our hypotheses are kept separate.
- 02
Map the buying questions
The questions your buyers ask at each stage, from first research to a quote or demo request. A question from a sales call is weighted differently from one a model invented.
- 03
Find where you are absent or unconvincing
Repeated, dated observations across search and AI answers, plus a review of your own pages and enquiry flow. Missing observations are recorded as missing, not as zero.
- 04
Assemble and approve the work
A backlog ranked against the commercial objective, priced against the execution allowance. Nothing is commissioned or published without your approval.
- 05
Execute with review gates
Agents prepare the work; people review it. Pages, product data, quote flows and technical fixes ship with versioning and rollback. Humans own commitments, sensitive claims and publishing.
- 06
Measure accepted enquiries, then re-plan
We reconcile observations, visits, enquiries and your team's accept or reject decisions every month, label what is observed versus estimated, and move the budget to what is working.
Discovery: the buying-question panel
Discovery is measured against a panel of the questions your buyers ask, balanced across four intent stages so it reflects the whole journey, not just branded search. Questions heard on sales calls are tagged apart from questions a model proposed, because only the first kind is demonstrated demand. A human reviews and tags every question.
Problem-aware
“how do I stop losing quotes to slower suppliers”
Buyers who feel the pain but haven’t named a solution or supplier yet.
Solution-aware
“commercial generator hire for construction sites”
Buyers comparing options, where being named and described accurately matters most.
Brand-aware
“is [your company] any good / pricing / reviews”
Buyers already considering you, where accuracy and evidence decide the enquiry.
Competitor-comparison
“[competitor] vs [your company]”
High-intent head-to-heads that AI now answers directly.
What we extract from every answer
A parser agent reads each response and emits structured data. This is classification, not generation, so it is cheap, fast and consistent.
Mention rate
How often you appear, per engine and per question category.
Representation
Whether what the engine says about your offer, regions and capacity is correct.
Position in answer
First mention versus buried at the bottom of the response.
Recommendation strength
Named as the top pick versus listed as an also-ran.
Citation sources
The URLs each engine pulls from to build the answer.
Competitor mentions
Who shows up alongside or instead of you, and how strongly.
The rules every reading is held to
Missing is missing, not zero
When an engine blocks sampling or a run fails, that cell is recorded as no reading. It is excluded from the denominator, never counted as invisibility.
A versioned question panel
Every run records the panel version, engine, route, locale, search configuration and date. When the panel changes, the trend line restarts rather than silently continuing.
API and consumer interface are labelled apart
An answer fetched through an API is not the same as what a person sees in the app. We label which route produced each observation.
Search Console is not an AI funnel
Google counts traffic from AI Overviews and AI Mode inside its overall Web reporting. We do not present that data as a cleanly separated AI-acquisition channel.
Bing’s AI reports are observational
Bing Webmaster Tools’ AI Performance reports sampled grounding queries: retrieval activity, not a log of your customers’ original questions. Its citation share is observational, not traffic share or a ranking score. We use it as an input, never as a result.
Correlation is labelled as correlation
“Accepted organic enquiries rose after these changes, but causality is uncertain” is a legitimate finding. We report it that way rather than inventing precision.
Why we sample
Single samples lie, so we repeat them.
The same question returns different answers run to run. We ask each question several times per engine and report the pooled result, with blocked or failed readings excluded rather than scored as zero. The panel is sized to the mandate: a published snapshot uses about twelve questions on whichever engines could be read on the day; a diagnostic or programme uses a larger panel scoped to the buyer journeys in the mandate. We publish our own runs, small panels included, on our numbers.
Published snapshot~12 questions
Diagnostic / programmesized to the mandate
Enginesthose readable on the day
Samples / questionrepeated
No readingexcluded, not zero
Headline outputaccepted enquiries
Want this pointed at your buyers?
Start with a free, capped snapshot of one buyer journey. If there is a commercial problem worth funding, we scope how to measure and win it.