Spectras

How is your brand doing in the AI-mediated web?

Buyers now ask ChatGPT, Claude, Gemini and Perplexity for recommendations before they ever reach your site. Prisma measures whether your brand appears in those answers, whether it gets cited as a source, and how often a competitor is named instead.

Open the dashboard How it works

Internal tool · single operator · a Spectras product

The mark is a citation matrix. Saturated cells are sources cited often; neutral cells are absence.

What it measures

Three questions, answered with evidence

Mention

Does the answer name your brand at all? Matched on name and domain, so a paraphrase still counts and a lookalike doesn't.

Citation

Did the engine cite your domain as a source? Being named and being sourced are different problems with different fixes.

Share of answer

Of every brand named in the answer, how much of it was you. Deliberately not “share of voice” — that means share of media spend, and this is a different number.

How it works

Repeatable by construction

An audit is only useful if you can run it again in three months and trust the difference. Every stage is built so a change in the score means a change in reality.

1
Define the clientDomain, competitors, the categories and audiences that matter.
2
Draft a prompt setBuilt from categories × audiences × intents. The same profile always drafts the same set.
3
Lock itA locked set is immutable. Editing one creates version n+1, so an old run keeps meaning what it meant.
4
Run itFanned across engines with repeats, capped by budget before a cent is spent, every answer stored forever.
5
Score and deliverDeterministic scoring — instant, reproducible and with no external calls — out to a branded scorecard.

Why the numbers hold

Evidence, not assertions

Raw answers kept forever

Every response is stored immutably — the database itself refuses to rewrite one. A finding can always be traced to the answer it came from.

Models frozen onto each run

A re-measurement reuses the models the previous run used, so score movement can't be confused with model drift.

Spend capped before it happens

Each engine has a hard monthly ceiling. A run that would exceed it is refused up front, not discovered halfway through.

Ready when you are.

The dashboard is where clients, prompt sets, runs and scorecards live.

Open the dashboard