# [Generative engine optimization agency: how to brief one](https://rebilder.com/solutions/generative-engine-optimization-agency)

> If you are hiring for GEO, the brief decides the outcome. What to specify, what to measure, and what no agency can promise however the contract is written.

- **Updated:** 2026-08-17

We are not an agency. If you are about to hire one for generative engine optimization, this is the brief we would want to receive, written from the side that builds the tooling.



## What you get

- A brief that specifies outcomes instead of activities
- The baseline to take before anyone starts
- Acceptance criteria you can check without the vendor
- The one clause that protects you if the engagement ends

## What the brief should specify

Weak briefs ask for activity ("optimise our site for AI search"). Strong ones name a measurable end state, and every item below can be checked by you afterwards without the agency’s cooperation.

1. **Name the templates in scope.** Not "the site". List the page kinds: product, category, location, article, service, FAQ.
2. **Require a baseline first.** A readability score per template against a published spec, taken before any change and handed to you as numbers.
3. **Specify the serving outcome.** `Accept: text/markdown` returns the page’s facts, and the response is a stated fraction of the HTML size.
4. **Specify fact position.** The first substantive fact appears within the first few lines rather than tens of thousands of characters in.
5. **Require the code to live in your repository.** This is the clause that matters most, and it is covered below.
6. **Name the reporting source.** Your server logs. Not a page-view tool, which cannot see agent traffic at all.

## The clause that protects you

Require that the serving-layer change lives in your own codebase and can be operated without the agency. This is not a hypothetical concern: if the change is a proxy the agency runs, then ending the engagement reverts every result you paid for, and the leverage that creates is not in your favour.

> **This is possible because the tooling is free** The gateway SDK is open, unmetered and self-hostable, and the readability spec is published with its arithmetic. There is no capability an agency can only give you by keeping the code, which means there is no honest reason for them to.

Acceptance check, run it yourself

```bash
# Facts present, and dramatically smaller than the HTML?
curl -s -H 'Accept: text/markdown' https://example.com/page | head -20
curl -s https://example.com/page | wc -c

# And Googlebot still gets canonical HTML?
curl -s -A 'Googlebot' https://example.com/page | head -5
```

## What to strike from any proposal

- **Guaranteed mentions or citations.** Assistants are not deterministic and no vendor controls their output. A guarantee here is unenforceable and the surest sign to keep looking.
- **An average position in AI answers.** There is no position in a generated answer to hold, so any number reported for it was invented.
- **Share of voice against named competitors**, unless those competitors opted in. Publishing that about a third party is not something we will do, and it is worth asking why anyone else will.
- **Results reported from analytics.** Agent fetches run no JavaScript and fire no beacon, so that report cannot contain what it claims to.

## What this does not do

- We are not an agency and take no engagements. Nothing here is a referral, and no provider is recommended on this page.
- It cannot make an assistant mention you. Assistants are not deterministic and no vendor controls their output, so a guaranteed citation is not a thing anyone can sell.
- It does not change what Google sees. Content negotiation returns the same substance in another format, and search crawlers keep receiving canonical HTML.
- We do not rank or review agencies. Nothing about a named third party is published without that domain owner’s verified opt-in, and that floor is not purchasable.

## What does a generative engine optimization agency do?

A technical half covering how your pages are served to AI agents, and an editorial half covering content that answers real questions. The technical half is a one-time change; the editorial half is continuous.

## What should I put in a GEO brief?

Name the templates, require a baseline before any change, specify the serving outcome and the fact position as checkable end states, require the code to live in your repository, and name your server logs as the reporting source.

## How do I check the work was done?

Request a page with `Accept: text/markdown` and read what comes back, measure it against the raw HTML size, re-run the readability scan against the baseline, and confirm Googlebot still receives canonical HTML. All four are free.

## What can no GEO agency promise?

That you will be mentioned or cited by a given assistant. Nobody controls model output, so that promise cannot be kept regardless of price or contract wording.

## Are you a GEO agency?

No. We build the gateway and the published readability spec that this work runs on. The scan is free, the SDK is free and self-hostable, and there is no service to buy from us.

## Related

- [Generative engine optimization services](https://rebilder.com/solutions/generative-engine-optimization-services)
- [GEO vs SEO: what actually changes](https://rebilder.com/learn/geo-vs-seo)
- [AI SEO agency: what they do](https://rebilder.com/solutions/ai-seo-agency)
- [AEO services](https://rebilder.com/solutions/aeo-services)
- [The published readability spec](https://rebilder.com/spec/ars)