14-DAY GEO BASELINE · BY APPLICATION
Prove one thing in 14 days: a wrong AI claim about your brand can be found, verified, fixed, and retested
Fixed scope, fixed deliverables, fixed boundaries. Ten working days from frozen prompts to a scheduled like-for-like retest — with every conclusion traceable to a snapshot, a source, and an approved fact.
01Who it fits
- B2B SaaS, developer tools, security, and other complex products where a wrong description has a real cost;
- Teams with an English site, a content owner, and the ability to publish changes within 14 days;
- A named person — product, legal, security, or marketing — who can approve facts within 48 hours;
- Teams who accept that retests are directional evidence, not ranking guarantees.
Not a fit: brands that want "rank #1 in ChatGPT," cannot modify public materials, or operate in high-risk verdict domains (medical, financial, legal advice) which we do not serve in this release.
02Fixed scope
| Item | Scope |
|---|---|
| Brand / market / language | 1 brand · 1 target market · 1 language |
| Measurement surface | 1 primary surface (authorized API; provider and model recorded per snapshot) |
| Prompt set | 10–15 real buyer questions, version-frozen before sampling |
| Sampling | 3–5 independent runs per prompt; natural and citation panels kept separate |
| Facts | 10–20 facts required for this round, each approved by your fact owner |
| Remediation | Up to 3 remediation tasks with fact-constrained drafts |
| Retest | 1 early retest + 1 scheduled follow-up, like-for-like |
03Ten working days
| Days | Work |
|---|---|
| D1–D2 | Brand materials, facts, and prompt set confirmed and frozen |
| D3–D4 | Baseline sampling; immutable snapshots archived |
| D5–D6 | Human review and client confirmation of findings |
| D7–D8 | Remediation tasks and fact-constrained drafts |
| D9–D10 | Publishing registration; retest run or scheduled |
Retest windows follow real crawl and platform behavior. The protocol does not promise visible change within 10 days — a null result is a valid, reported result.
04What we use, and what stays human
Two data classes only: your public brand materials (authorized by you) and answers returned by the declared measurement surface. Raw responses are stored immutably. Web content is treated as untrusted input; analysis runs without credentials or publish permissions.
Every client-visible conclusion is human-reviewed before you see it. Models only propose candidates. Opinions and complaints keep their attribution — we never adjudicate whether a reviewer was right.
We do not collect customer lists, chat logs, contracts, credentials, or non-public commercial material.
05Deliverables
The engagement delivers an approved question scope, a customer-facing comparison of observed answers and public brand statements, recommended public content changes, and a like-for-like recheck summary. See the sanitized public demo case.
06What we don't promise
- No ranking, recommendation, or mention-rate guarantees on any platform;
- No claim that API measurements equal what every consumer sees in ChatGPT, Gemini, or Perplexity;
- No automated edits to your site; no publishing on your behalf;
- No adjudication of user complaints or reviews;
- No "proprietary AI score" — you get n/N per surface with controls and limits.
07How to apply
The application form opens once the operating entity, responsible contact, data region, and response SLA are published — we don't put placeholder forms or fictional inboxes on this page.
Meanwhile: read the public demo case to see the customer-facing result, and the baseline protocol for the sampling and review discipline.