Avouch

//.the_platform

The Citation Intelligence Platform.

An instrument, not a brochure. Before we move a number, we measure it — the same way, every time, across every engine.

Start free. No commitment.

Cited in ChatGPT, Claude, Gemini, Perplexity, Grok, Brave, Google AIO, Copilot.
//.the_sweep_001
FIG. 01The seven-engine sweep
ChatGPT
Claude
Gemini
Perplexity
Grok
Brave
Google AI Overviews

+ Copilot snapshot

Every Recommendation Index is built from the same declared set of buyer questions across seven engines — and a Copilot snapshot — in one sweep. No tool swapped mid-flight, so the readings are comparable to each other and to next month's.

//.a_real_reading

Not a mock-up. A real client’s reading

One sweep, one real client, name withheld. This is what the platform actually returns — not the illustrative sample below it.

26.7%mentioned in AI answers
44.5%share of voice
25.6%answers citing their site
7,796answers analysed, 7 engines

Mention rate by AI platform

Grok39.2%ChatGPT28.4%Gemini27.7%Claude26.2%Perplexity23.5%Brave20.1%Google AIO13.7%
Each engine keeps its own sample size — rates are never blended across engines.
Verified client result · anonymised

Sentiment on the answers that do mention them: 57% positive, 43% neutral, and a single negative mention out of 855 sentiment-scored answers — read, sourced, and shown to them with the exact quote.

Avouch console · sentiment · negative mentions
The platform's negative-mention panel: the one negative sentiment answer out of 855, shown with its exact source prompt, engine and confidence score
That negative mention, in full — client name blacked out, nothing else touched.
Verified client result · anonymised
//.the_readouts_002

Six readings, one honest number

Each sweep returns more than “are you mentioned”. Your Recommendation Rate™ — how often you’re the answer, per engine — position-weighted share of voice, a whitespace map of every gap, a ranked list of what to fix first, and a check on whether the engines are telling the truth about you.

FIG. 02Position-weighted share of voice
You31%Rival A22%Rival B18%Rival C14%
Share of AI answers naming each firm, weighted by where it lands in the answer. Illustrative.
Illustrative example — not a real client

The whitespace grid — where you're cited, and where a rival owns the answer:

Query clusterChatGPTClaudeGeminiPerplexityAIOGrok
Best wealth firm, SurreyCitedPartialAbsentCitedAbsentPartial
IFA for business ownersPartialAbsentAbsentPartialAbsentAbsent
Multi-generational planningCitedCitedPartialCitedPartialAbsent
Alternatives to the big namesAbsentAbsentAbsentPartialAbsentAbsent
CitedPartialAbsent
Per query-cluster × engine, across one sweep. Illustrative.
Illustrative example — not a real client
//.what_to_fix_003

The work, ranked by value

The platform doesn't just show the gaps — it ranks them, so the first month's work is the highest-value citations to win, not the easiest ones to tick off.

  1. Multi-generational planning explainer

    A rival owns this answer on four of seven engines. One citable, passage-level page is the fastest way in.
  2. “Surrey + business owner” queries

    You're partial on two engines — a cleaner, front-loaded answer turns a mention into a citation.
  3. Independent-review presence

    Here the engines lean on VouchedFor and Trustpilot; your profile is thin, so the citation goes elsewhere.
Illustrative example — not a real client

The hallucination battery

Fifteen brand questions, run across the engines, flagging where an AI states something untrue about your firm — the risk no ranking tool thinks to look for.

//.why_it_holds_004

Why the number holds up

1,300automated tests, done before every Recommendation Index ships
36,000+ AI answers analysed
The evidence base behind the Recommendation Index — real AI answers logged across seven engines plus a Copilot snapshot, and growing with every sweep. Not estimates.
Declared, versioned prompt universe
The exact questions are written down and version-controlled — so the number is reproducible and auditable, not a one-off you have to take on faith.
Parametric baselines
We measure what each model already knows about you before any retrieval — the floor every later reading is measured against.
K-run averaging
Every figure is an average of multiple runs, so it's statistically stable — not a single lucky (or unlucky) snapshot.
Multi-turn + sentiment
Not just whether you're named, but in what light, and across a real back-and-forth — the way buyers actually ask.
//.how_it_runs_005

How a sweep runs

  1. Define

    We build the declared prompt universe for your sector — the real questions your buyers put to the engines, version-controlled from day one.
  2. Discover

    The sweep runs across seven engines, with a parametric baseline of what each model already knows about you before any retrieval.
  3. Analyse

    Position-weighted share of voice, the whitespace grid, the hallucination battery, and a value-ranked list of the gaps to close first.
  4. Optimise

    We produce the citable work — then re-run the identical sweep, so you see what moved, and when.
//.how_it_differs_006

Reproducible, not vibes

Most AI-visibility tools hand you a number you can't reproduce. The CIP is built so you can re-run it — and so the work that moves it lives in the same place.

DimensionAvouch CIPTypical AEO tools
The numberReproducible — declared prompts, averaged over runs, re-run on requestA single snapshot you can't reproduce
The promptsA declared, versioned prompt universeAn ad-hoc handful, changed between reads
The pipelineOne pipeline we own, end to endVendor tools swapped mid-engagement
The baselineParametric — what the model already knowsStarts at retrieval; no baseline
After measuringWe produce the work that moves itMeasurement only — the work's on you
How the CIP compares to a typical AEO tracking tool.

//.the_standard

If we can’t re-run the number in front of you, it doesn’t go in the deck.

The platform exists so every figure on every page traces back to a sweep you could watch us run again.

The Avouch measurement standard

//.the_point_007

Measurement is the floor. Being vouched for is the point

All of this rigour serves one outcome: the AI names you, and has good reason to.

Showing up

visible · mentioned · on the list

the vouch

Vouched for

the client who chose you

//.questions

Platform questions

Is the platform something I log into?
No — it's how we produce your Recommendation Index and run your retainer, not a self-serve dashboard. You get the readings and the work; we run the instrument. A client view may come later.
Which engines do you measure?
Seven — ChatGPT, Claude, Gemini, Perplexity, Grok, Brave and Google AI Overviews — plus a Copilot snapshot. More are wired and switched on as they start to matter.
What makes your number more trustworthy than another tool's?
It's reproducible. The prompts are declared and version-controlled, every figure is an average of multiple runs, and we own the pipeline end to end — so we can re-run the exact measurement and get the same answer.
Are the sample numbers on this page real?
One reading is: the mention-rate figures are a real client's actual sweep, name withheld, tagged as verified. The share-of-voice, whitespace and opportunity examples further down are illustrative and clearly tagged as such. Your Recommendation Index returns your own real figures either way.

AI can't recommend what it doesn't know.

See where you stand

Start free. No commitment.