Should an AEO agency guarantee results?

An AEO agency should not guarantee results, and you should not want it to. A guarantee moves delivery risk onto the supplier, which sounds like a win until you notice that the supplier then chooses which number it is on the hook for, and it will choose the cheapest one to move.

The distinction that matters is between outcomes and deliverables. An outcome is being named when a buyer asks ChatGPT which vendors to consider, and it depends on an engine's retrieval decision that nobody outside that engine controls. A deliverable is twelve rewritten pages, a documented prompt set run weekly, and a monthly report whose method is written down. The second is guaranteeable because it is work. The first is not, because it is somebody else's judgement.

Where a supplier does offer an outcome guarantee, the wording is where the answer hides. A promise of a named placement or a percentage visibility lift inside a fixed window can only be honoured by controlling the measurement, and picking enough low-competition prompts will raise almost any number. The guarantee gets paid and the pipeline does not move.

Our own position on this is on the record, which is the fairest way to put it: we do not guarantee citations and we say so in writing. What we commit to is the method, the cadence, the people and the exit terms. Our page on how to choose an AEO agency treats a citation guarantee as a reason to walk away, and this page is the longer argument for why.

Why can nobody guarantee an AI citation?

Nobody can guarantee an AI citation because three separate things stand between a good page and an answer, and a supplier controls one of them. The page is the one you can buy. Retrieval and generation are not for sale.

The primary source is unusually blunt about this. Google's documentation on AI features and your website states that there are no additional requirements to appear in AI Overviews or AI Mode and no special optimisations necessary, and then says that meeting all requirements and best practices does not guarantee that Google will crawl, index or serve your content. Indexing and serving are not guaranteed. That is the platform telling you, in its own documentation, that the thing an outcome guarantee promises is outside anyone's gift.

The research points the same way with numbers attached. Chen and Koudas, analysing 158,099 answers and 468,037 citations across two 90-day collections in 2026, found that which pages an engine's own retrieval returned explained about 32% of the variation between pages, against about 2.8% for the properties of the pages themselves. They also found the repeatable part of that variation drifts over weeks, and that models write brands they already know into their own sub-queries.

Say the honest version of that in both directions, because it cuts at us too. A page score measures whether a page is in a condition to be retrieved and used. It does not predict which engines will cite it, and the measurements above say nobody should claim it does. Anyone guaranteeing a citation is claiming to control the 32% in that study as well as the 2.8%, and no supplier does.

What can an AEO agency legitimately commit to?

An agency can commit to anything inside its own control, and the list is longer than most contracts bother with. The table separates what can be guaranteed from what cannot, and what to accept in place of each promise that should not be made.

What a supplier might promise	Can it be guaranteed?	What to put in the contract instead
Named in ChatGPT for your category	No. The retrieval and generation decisions sit with the engine	A stated target prompt set, with monthly reporting of which of your URLs appeared as cited sources
A percentage increase in AI visibility	No, and the metric is unstable enough that the promise is close to meaningless	A baseline both sides sign off, reported with an interval, on a prompt set that cannot be changed unilaterally
Pages delivered and rewritten	Yes. Pages are a deliverable the supplier controls	A monthly count, with a definition of what counts as delivered and who approves it
Measurement cadence and method	Yes	Prompt count, runs per prompt, engines covered, and the formula written out rather than described
Who actually does the work	Yes	Named people, their time commitment, and notice if the team changes
What you keep if you leave	Yes, and this is the clause most often missing	The pages, the prompt set and the historical measurement data, in a portable format, with no notice penalty

Read the bottom row as the real test of a supplier's confidence. A firm that expects to earn renewal on results does not need to hold your measurement history hostage, and a firm that does need to is telling you what it thinks its results will look like. Our page on switching AEO agencies without losing progress covers what portability should mean in practice.

What does accepting a guarantee actually cost you?

Accepting a guarantee costs you the choice of metric, and the metric is the whole engagement. Once a supplier carries delivery risk against a number, that supplier has a strong interest in which number it is, and you have handed over the decision that determines whether any of the work was worth buying.

Three substitutions follow, and all three look like success on a dashboard. A prompt set weighted towards questions you already win, so the baseline rises without new ground. Low-competition phrasings nobody buys from, which are cheap to capture and worth nothing. And mention counts on branded questions, where an assistant naming you in answer to a question about you is not evidence of anything.

The instability of the underlying measurement makes all three easier to hide. Schulte, Bleeker and Kaufmann ran four AI search engines daily across four verticals for 45 to 46 days from 24 January to 20 March 2026 and found cited sources overlapping only about 34% to 42% between consecutive days, and about 32% to 43% between repeated runs inside a single 24 hour window. A number that moves that much on its own will produce a rise somewhere in any reporting period, and a guarantee written against it can usually be satisfied by waiting.

What you give up by refusing the guarantee is less than it seems. You keep the right to choose the questions that matter commercially, which is the only part of the measurement that connects to revenue. Our page on how to tell whether your AEO is working covers which movements are worth believing.

How should a contract handle measurement when the numbers move on their own?

A contract should fix the method and leave the number free to move, which is the opposite of how most retainers are written. Fixing a target number against a drifting measurement produces an argument every quarter. Fixing the method produces evidence.

Four clauses do most of the work. A baseline measured and signed off before any work starts, so there is an agreed starting point. A frozen prompt set that neither side can change alone, with a documented process for adding prompts as the category moves. A stated run count per prompt, because a single run per month is not a measurement. And reporting that carries an interval rather than a point, since a share estimate at 95% confidence needs roughly 100 observations to support a plus or minus 10 point interval and roughly 400 for plus or minus 5.

Then add the clause almost nobody includes: what counts as a result. Decide in advance whether you are buying appearances in answers, cited source placements, referral sessions or pipeline, and write down which one the renewal conversation will turn on. Teams that skip this spend month nine arguing about which chart matters, and the supplier usually wins that argument because it chose the charts.

Track record is the right substitute for a promise, provided it comes with its limits attached. Our own early results for law firms are published as two anonymised case studies with an honest account of what the numbers do and do not measure, which is the form a track record should take. A supplier willing to tell you what its measurement cannot prove is giving you more information than one offering a guarantee.

What should you ask for instead of a guarantee?

Ask for five things instead of a guarantee, and ask for them before price. Each one is answerable in a sentence, and the quality of the answers separates suppliers faster than any proposal document.

Ask what the method is, in enough detail that you could run it yourself. Ask for the baseline before work starts, signed off by both sides. Ask what you keep if you leave, specifically the pages, the prompt set and the measurement history. Ask who does the work and how much of their time you get. And ask what would make them tell you to stop, because a supplier with no stopping condition has no definition of failure.

That last question is the one that separates them. An honest answer sounds like a condition: if the assistants are not producing vendor answers in your category after a quarter of measurement, there is nothing to win yet and the retainer should stop. Our page on when to stop paying for AEO sets out what those conditions look like from the buyer's side, and whether AEO works when nobody is searching your category covers the case where the answer is not yet.

Where a supplier insists on offering a guarantee, read the definition rather than the headline, and check which metric it pays out on. A guarantee against a number the supplier selects, measures and reports is not a guarantee. A free visibility assessment will give you an independent baseline to hold any proposal against, including ours.

What else do people ask about AEO guarantees?

The five questions below about guarantees come up while a retainer is going through procurement, usually when legal asks what the supplier is actually promising.