What We Promise GEO Clients in 2026, and What We Never Promise
Why no agency can guarantee a ChatGPT citation or position, what a GEO agency can commit to instead, and how we write reports so clients can see cause and effect.
Sooner or later in every sales call, someone asks it straight: can you get us into ChatGPT? The honest answer is that we can make it much more likely, and we can't guarantee it. Neither can anyone else. An agency that guarantees a citation, or a position in an AI answer, is selling a screenshot.
We'd rather lose that deal than win it on a promise we can't keep. But "no guarantees" on its own is a weak pitch, so we pair it with a specific list of what we do commit to and a reporting format that lets the client check us. This post covers both.
Why a citation can't be guaranteed
An AI answer isn't a results page with fixed slots. It's generated fresh, and four things change it.
The run. Ask the same model the same prompt twice and you can get different brands, in a different order, citing different sources. One check proves very little in either direction.
The location. Answers differ by country and often by city. Google AI Overviews are region-specific. A brand can be the first recommendation for a prompt in Amsterdam and missing for the identical prompt in Utrecht.
The persona. A finance director and a first-time buyer phrase the same need differently, and the model responds to those differences. Persona tracking exists because a brand can be visible to one buyer type and absent for another.
The model. ChatGPT, Gemini, Perplexity, Microsoft Copilot, Google AI Mode, and Claude each pull from different sources and weigh them differently. Winning one tells you little about the others.
On top of all that, the models change. Sources that were cited every day last month can drop out after an update nobody announced. So a "position" in an AI answer is an observation from one model, one place, one persona, one moment. Anyone who promises a fixed one is either measuring a single convenient prompt or not measuring at all.
What we do commit to
We can't control the output. We can control the inputs and the honesty of the measurement, and we put those in writing.
Process and cadence. The prompt set is signed off in the first month and reviewed every quarter. The monthly report arrives on an agreed date. The weekly site audit gets read every week, not when something breaks.
Prompt coverage. The scope lists exactly which prompts we track, on which models, in which markets, for which personas. Coverage doesn't quietly shrink when results are bad, and new prompts go in as a separate group so the trend stays comparable.
Crawl access. Every crawl error on a priority page gets fixed or escalated with a named owner. We can't make an AI system cite a page, but we can make sure its crawlers can fetch it, and we report the status every month.
Pages shipped. An agreed volume of work each month, logged with dates and URLs, including what's still waiting on the client.
Honest outcome reporting. Citations, crawls, and AI-referred visits, reported every month, down months included.
That last one is a promise about reporting, not about results. It's the one clients test hardest, usually in the first month a number drops.
Setting expectations with difficulty scores
The most useful expectation-setting tool we have is the difficulty score attached to each tracked prompt. It tells us how competitive a prompt is before we've done any work.
Before tracking starts, we sort the client's prompt set into difficulty bands and show the split. Lower-difficulty prompts are where we tell clients to look for early signs of movement. High-difficulty prompts usually depend on third-party sources that AI systems already trust: publications, Reddit threads, YouTube videos, comparison pages. Moving those takes offsite work and more time, and we say so at sign-off rather than in month five.
It changes the conversation. A client whose prompt set is mostly high-difficulty knows from day one that the early months are about access, owned pages, and offsite groundwork. Nobody is surprised, and nobody reads a slow quarter on hard prompts as failure.
Reporting so the client can see cause and effect
A visibility chart going up or down doesn't tell anyone why. Our reports are built around the chain from a change we made to a result we can observe, one page at a time.
For every page we ship or change, the report carries a row with:
- The date it shipped.
- The first AI crawler fetch after the change, and from which operator.
- The first citation of that page in a tracked answer.
- Its ramp-up days and peak rate, as Citation Trends reports them.
- Its current citation rate, and whether it's past peak.
Promptwatch's Agent Analytics supplies the crawl side, including the crawl-to-citation path that connects a fetch to the citations that followed. Citation Trends supplies first cited, ramp-up days, peak rate, and decay half-life per cited page. When we need to watch one URL closely, Page Tracker follows it across prompts. Visitor analytics adds the AI-referred visits that page received.
Read together, the row tells a story the client can verify. We changed the page on this date, OpenAI's crawler fetched it on that date, it was first cited after that, and it took this long to ramp. When the chain breaks, the row shows where. A page that was never fetched again is an access problem. One that was fetched and never cited is a content or authority problem. Those need different work.
We add one guard against false credit. If every tracked prompt moves on the same day, a model change is a likelier explanation than anything we shipped, and the report says that too.
Decay half-life also feeds back into what we promise. A page whose citations fade fast after their peak is due for a refresh sooner than one that holds. That turns the refresh schedule into evidence instead of a calendar habit.
The sentence in every proposal
Every proposal we send has a version of this line: we don't guarantee placement in any AI answer; we guarantee the work, the cadence, and a report that shows what changed and why.
Some prospects find that underwhelming next to a competitor's guarantee. We let them compare. A guarantee with no measurement behind it can't be checked, and a commitment list with a monthly report can be checked line by line.
Making the commitments measurable
Promises about process only mean something if the client can check them. We run client programs on Promptwatch because it turns each commitment into something visible: prompt tracking with difficulty scores, personas, and locations for coverage; Agent Analytics error tracking and the crawl-to-citation path for crawl access; Citation Trends ramp-up days and decay half-life for the effect of each page we ship; visitor analytics for AI-referred visits. The scheduled PDF report puts it in front of the client every month whether the numbers are good or not. If you want a program scoped this way, write to hello@1001seomedia.com.