FOR AGENCIES AND MULTI-BRAND TEAMS

PROVE AI VISIBILITY
WITH EVIDENCE

A score tells a client where they stand. An evidence pack shows them why, and holds up when they ask. Fixed prompts, the raw answers from ChatGPT, Claude, Gemini and Perplexity, and the same competitors measured on the same questions.

FIXED PROMPTS · RAW ANSWERS · THE SAME COMPETITORS ON THE SAME QUESTIONS
CHATGPT · CLAUDE · GEMINI · PERPLEXITY · DEEPSEEK · LLAMA · MISTRAL · KIMI · MINIMAX · GROK

Why a score alone does not survive the meeting

A visibility score is one number standing in for hundreds of answers. It is the right thing to open a client review with and the wrong thing to defend one with.

The number arrives without its working

Somebody asks which questions produced it. If the only answer is a dashboard, the conversation turns into trust rather than evidence.

The prompts move between runs

Change the questions and the score changes with them. Without a locked set, this month and last month are not comparable and neither is the story you tell about them.

The work leaves with the person who ran it

When the account changes hands what transfers is a slide. The reasoning stays with whoever did it.

What counts as evidence

A pack is complete when somebody who was not in the room can repeat the run and get the same shape of answer.

A fixed prompt register

The questions are written down per client and do not change between runs without a note saying why.

A stated cadence

Weekly, monthly, or before each review. The period is labelled on every result.

Raw answers, not summaries

The full response each platform gave, kept as it came back.

Mention, citation and recommendation kept apart

Being named, being linked and being put forward are three different outcomes, and a pack counts them separately.

The same competitors on the same questions

A comparison means something only when both brands were asked the same thing on the same day.

Platform and period on every figure

Every number carries which engine produced it and when.

The method

Six steps an agency can run, sell and hand over.

  1. Lock the prompt register

    Write the questions a buyer would actually put to an assistant. Agree them with the client and freeze them.

  2. Choose platforms and cadence

    Decide which engines matter for this brand and how often to ask. A smaller set asked often beats every engine asked once.

  3. Capture the raw answers

    Keep the full response rather than a summary of it. The answer is the evidence.

  4. Tabulate against defined metrics

    Score with the definitions written down, so the next person scores the same way you did.

  5. Compare named competitors on the same prompts

    Same questions, same day, same engines. Anything else is two studies wearing one cover.

  6. Package it for the review

    Cover, method, prompts, results by platform, competitor table, caveats, next actions.

The words, defined

Agree these with the client before the first run, not after the first challenge.

Mention

The brand appears in the answer by name.

Citation

The answer links to the brand's own page as a source.

Recommendation

The answer puts the brand forward as an option rather than naming it in passing.

Presence rate

The share of prompts in the register where the brand appeared at all, on one platform over one period.

Share of voice

How often the brand appeared against the named competitor set, on the same prompts.

Recommendation rank

Where the brand sat in the order when an answer listed options.

What a client pack contains

An example structure rather than a report Eclipse has published.

  1. Cover

    Client, period, platforms, and who ran it.

  2. Method

    The prompt register, the cadence, and what was counted.

  3. Results by platform

    Each engine on its own, because they do not agree with each other.

  4. Competitor table

    The named set, measured on the same prompts.

  5. Caveats

    What the run could not see, and what changed since the last one.

  6. Next actions

    The gaps worth writing against, in the order worth writing them.

Where Eclipse fits, and where it does not

Some of this is product and some of it is how an agency chooses to work. The split matters when you are selling it.

GEO profiles hold the register

Generative engine optimisation profiles carry the buyer questions, the competitor brands and the platforms to scan. That is the locked set the method asks for.

A GEO profile in Eclipse, with per-platform performance across ten engines and the questions it runs

Runs are weekly or on demand

Profiles run on a weekly schedule or when you run them, and usage follows the platforms you pick.

The recent executions table on a GEO profile in Eclipse, each run stamped with its platform, date and sentiment

Logs keep the raw answers

Each run is recorded by profile, platform, date, status and mention count, and opens to the full response so you can copy it into a client pack.

A run in Eclipse opened to the full answer Claude gave, with the sources it drew on listed beneath it

Competitors are named and discovered

You set the competitor brands. Eclipse also surfaces others that keep appearing alongside the client.

The competitive co-occurrence list in Eclipse, naming every brand that appeared alongside this client and how often each was co-cited

Client workspaces stay separate

Every client is its own workspace, so one client's register and history never appear inside another.

One client workspace in Eclipse, holding that client’s own positioning, brand, proof and trusted sources and nothing from any other

The pack itself is yours

Eclipse holds the runs, the answers and the history. Assembling them into a client document, and the language used in it, is agency work.

Common questions

PROVE IT ON A CLIENT YOU ALREADY RUN

Bring one account and a handful of the questions their buyers ask. We will set the register up on the call.