Selling GEO as an agency service without promising a place in the answer

Your own agency is the first case a prospect checks

A prospect who hears that you now sell GEO can test the claim in a minute: ask an assistant to name agencies that do what you do, in your city, for clients like yours. Whatever comes back is your first case study, written by nobody at your agency. It is also the only one a prospect can run without asking you.

So measure yourself before you sell. Build a panel from your own services, client industries and budget bands (the grid is in AI visibility for marketing agencies), run it, keep every answer, and date the reading. Then make the changes the page for marketing agencies lists for an agency's own site, and run the same panel again on a date you set before the first change.

The two dated readings and the answers behind them are the first evidence you own outright. They show that you run the method on yourself, and what moved after your own changes. They are one agency's result, not a forecast for a client. Say so when you show them; a prospect who has been told the limits can hold you to the rest.

A low first reading is no reason to wait. The probe in our audit of 9,483 small business websites gives a sense of the starting point: across 40 trades, one buyer question per business went to Claude with web retrieval on, and 14.9% of the 9,240 businesses were named, with no separate figure for agencies. A prospect can check a before and an after. A clean number with no history gives them nothing to check.

Three lines in the scope: the audit, the fixes, the monthly reading

Sell GEO as three lines, each with its own deliverable, price and end. They finish at different times, and a client may want one without the others: an audit with the fixes done in-house, fixes from an audit someone else ran, or the reading alone. Separate lines also keep the reading honest when the agency that made the fixes is the one measuring them, which is the subject of measuring AI visibility for agency clients.

What a real audit contains, part by part, is in the GEO audit, and it is not repeated here. What matters for the scope is that each line has a finish the client can see: a report walked through, a change live with its date, a reading on the agreed day.

How to price each line with your own figures

Price each line from what it costs you to deliver, and write the figures into the proposal as your own. The work is countable. A panel of twelve questions on four assistants at three runs each is 144 answers for a person to read, once for your audit's baseline and again every month after it; the crawl, the competitor pages and the fix list sit on top of that first reading.

Audit            [fixed fee], delivered [n] weeks after the panel is signed
Fixes            [fixed fee] for the scope the audit sets,
                 or [day rate] a day, estimated at [n] days
Monthly reading  [monthly fee], ended with [n] days' notice

The fix line is the hardest to price, because it depends on things outside the agency: whether you can edit the client's site or have to brief their developer, who writes and approves the copy, and how many services and locations the fix list covers. Say which of those the price assumes. A fixed price for a scope the audit has already set is easier to hold than an estimate made before anyone has read an answer.

For comparison, and only as one way to price these lines, ours are on the pricing page, for a narrower reading on one assistant, Claude, with one run per question: the free visibility check at $0, where a person sends a written result within 48 hours of receiving a website and an email address; the AI Visibility Audit at $490, one-time, in one week, credited in full against optimization; GEO Optimization from $1,900, for a fixed scope the audit sets; and Monitoring at $290 a month, which you can cancel in any month. Nothing on this page depends on buying them.

What the contract promises, and what it never promises

The contract promises work and readings, because those are what the agency controls.

• The audit, with the parts the scope lists, by [date].
• Each change on the fix list made or handed over by [date], and logged with the day it went live.
• A reading on the signed panel every month, with every answer attached.
• The panel, the answers and the log handed over complete if the client leaves.

It never promises a place in an answer, a position or a ranking, a number of mentions, being first, a named rate by a date, or a traffic or revenue figure. The reason fits in one sentence, and it belongs in the proposal: an agency can change what the pages say, and nothing it does reaches inside ChatGPT, Claude, Gemini or Perplexity, where the answer is written. Advertising an operator sells, where it sells any, is bought on the operator's terms, and it is a different thing from being named in an answer.

Google says much the same about its own features. Its guide to generative AI search says that, from Google Search's side, optimizing for AI Overviews and AI Mode is still SEO, and that no third-party tool has access to its internal ranking or AI systems. A promise of placement is also a claim about results, and the rules on such claims in proposals and advertising differ by jurisdiction; your counsel can say which apply to yours.

Some work stays out of the scope whatever a client asks for. Google's spam policies give pages targeted at specific regions or cities that funnel users to one page as an example of doorway abuse, and Google's guide to generative AI search says that separate content for every variation of how people search, made mainly to manipulate rankings or AI responses, violates the scaled content abuse policy. The guide also says seeking inauthentic mentions is not as helpful as it might seem, and Reddit's content policy forbids content manipulation such as spamming. None of it belongs on a fix list, and a brand mention an agency planted is not evidence of anything.

What to show a prospect, and what each piece proves

Show evidence a prospect can check, with its dates on it. Each piece proves something narrower than it looks, and saying what it does not prove is part of showing it.

Leave out any before and after measured on two different panels, and any sentence like "we got [client] into ChatGPT". The first compares two instruments. The second claims a lever no agency has.

When a client asks for a guarantee

Answer in three parts, in this order: what cannot be promised and why, what is promised instead, and how the client will judge the work. A request for a guarantee is also a question about how the client will know the money was well spent, and the third part answers it. In a proposal, the three parts can read like this.

What we promise
- [The audit], with the parts listed in the scope, by [date].
- [The fixes] on the audit's list, made or handed over by [date],
  each logged with the day it went live.
- A reading every month on the panel you signed: [n] questions,
  four assistants, [n] runs each, every answer attached.
- Everything above handed over in full if you leave.

What we do not promise
- A place, a position or a number of mentions in any answer from
  ChatGPT, Claude, Gemini or Perplexity. We can change what your
  pages say. Nothing we do reaches inside those systems.

How you will judge it
- After [n] monthly readings, we compare [the questions for your
  main service] with the baseline, against the spread between runs.
  If they have not moved, we change the work or stop, and you can
  end the reading with [n] days' notice.

Have counsel read the final wording before it goes into a contract. The shape matters more than the phrasing: every promise is something the agency does, and the one thing it cannot do is named plainly, once, beside them.