How do you know when to stop paying for AEO?
Three tests decide it. First, is the measurement real: same questions, same engines, raw answers attached, on a schedule. Second, has the gap closed: are you named across the question set, and has anyone been displaced. Third, is the remaining work maintenance or growth. A retainer that fails the first test should end now. A retainer that passes all three has usually earned a smaller scope rather than a cancellation, because citations rotate without anyone touching them: a Scrunch and Stacker analysis of 3.5 million citation events, reported by Similarweb's AI search team in 2026, put the median time for half a page's AI citations to drop off at about 4.5 weeks. Flat is therefore a result, and the question is what maintaining it is worth.
How do you know when to stop paying for AEO?
Stop paying for AEO when the measurement cannot be verified, when the scope has drifted into content production you did not ask for, or when your position is stable and nobody is contesting it. Those three endings are different, and only the first is an emergency.
Test the measurement first, because it is the cheapest thing to check and the most common failure. Ask for the raw answers behind last month's report. A supplier who can produce them within a day, matched to a question set that has not quietly changed, is doing the work. A supplier who sends a prettier summary instead has been selling you a dashboard, and no amount of patience fixes that.
Test the gap second. Take the ten questions that matter most commercially and read who the engines name. If your competitors appear and you do not, after two quarters of work, the engagement has not delivered its central promise and the conversation is about why rather than about renewal. If you are named across most of the set, the gap you hired someone to close is closed.
Test the remaining work third. Growth work means questions your site still does not answer, markets you have not covered, and competitors who outrank you in answers. Maintenance work means holding a position you already have. Paying growth prices for maintenance work is the most common way these engagements overstay, and it is a scope problem rather than a supplier problem.
Our guide to judging whether AEO is working covers the evidence each test needs, and whether AEO is a project or a retainer sets out which parts of the work genuinely recur.
What does "done" look like for AI search visibility work?
Done, for AI search visibility, means your site answers every question in the tracked set with a page the engines can read, you are named consistently in the answers that matter, and the remaining movement is other people's. Reaching that state is normal. Staying there without any effort is not.
Three conditions make a reasonable definition of done, and each one is checkable. Every tracked question has a page that answers it outright in its own first lines. Your pages appear as named sources across repeated runs rather than once. And where an answer lists vendors you are in the list, with the competitors who used to be there without you no longer unopposed.
What done does not mean is a fixed number you can defend forever. The engines re-select sources constantly, so holding a position is active rather than passive. Parse's volatility research, which re-ran more than 16,000 prompts around 22 times each across ChatGPT and Google AI Overviews in March and April 2026, found repeat answers to the same prompt shared only about a fifth of their cited sources on ChatGPT. Your name reappearing in most runs is the signal, and the specific pages behind it will keep rotating.
Treat done as a trigger to change the scope rather than to end it. The work that keeps a position is a fraction of the work that won it: the measurement continues, a few pages get updated when the log shows a loss, and the question set gets revisited when your product or market changes. Pricing that stays flat through that transition is the thing to renegotiate.
Does your AI visibility decay if you stop paying?
AI visibility does decay when the work stops, though more slowly than vendors imply and faster than most buyers expect. The honest answer is that nobody can give you a single number, because the published studies measure decay in different ways and arrive at different figures.
The most widely cited figure comes from a Scrunch and Stacker dataset of 3.5 million citation events collected between September 2025 and March 2026, reported by Similarweb's AI search team: a median of about 4.5 weeks for half of a page's AI citations to drop away, with ChatGPT around 3.4 weeks, Google's AI surfaces between 4.3 and 4.7 weeks, and Perplexity around 5.8 weeks. The same analysis reports that brands with active earned media held citations for about 10 weeks on average.
A shorter figure comes from a different method. Profound's analysis puts the half-life of an AI citation at 11 days, measured from each page's peak citation share rather than from its first appearance, and reports that 78% of page and engine pairings observed for at least two weeks past peak had a half-life of 14 days or less. Measuring from the peak produces a shorter number by construction, which is why the two studies are not directly comparable.
What both support is the same practical conclusion. Decay runs on a scale of weeks, so a quarter without measurement is long enough for a loss to happen and go unnoticed, and the first thing you lose when you cancel is not the citations but the ability to see them going. Our page on measuring AI citation decay sets out how to calculate your own half-life from a run log rather than borrowing anyone's benchmark.
Should you stop, scale back, or keep the full scope?
Choosing between stopping, scaling back and keeping the full scope depends on two things: whether the measurement is trustworthy, and whether anyone is contesting your position. The table sets out which situation points at which decision.
| Decision | The situation that justifies it | What you keep paying for | The risk you accept |
|---|---|---|---|
| Stop now | Reports cannot be traced to raw runs, or the question set keeps changing under you | Nothing beyond the notice period | None worth protecting, because the series was never comparable |
| Stop and take it in-house | Someone on your team owns organic search and can publish within a week | A tracking subscription and your own staff time | The monthly runs quietly stop happening when a quarter gets busy |
| Scale back to maintenance | Named across most of the question set, no active displacement, no new markets | Measurement, a short report, and a small fixed amount of page work | Slower response when a competitor starts publishing seriously |
| Keep the full scope | Questions still uncovered, or rivals being named where you are not | Measurement, rewriting, new coverage and technical work | Paying growth rates for a position that may already be defensible |
| Pause for a quarter | A migration, rebrand or product change makes page work pointless until it lands | The measurement runs only, so the series survives the pause | Losing ground during the pause without the budget to respond |
The pause row is the one most often got wrong. Cancelling the measurement during a replatform saves the smallest line in the budget and destroys the comparison you will need afterwards, which is exactly when you most want to know what changed. Our guide on whether a redesign or migration hurts AI visibility covers that window in detail.
Which parts of the work can you take in-house when you scale back?
The parts worth taking in-house are the repeatable ones: running the question set, logging the answers and updating pages your own team wrote. The parts that resist being internalised are the judgement calls, which is reading a flat month and deciding whether it is noise, a competitor or a technical fault.
Measurement is the easiest to move and the first to be dropped. Running 20 questions across four engines and pasting the answers into a sheet takes an afternoon a month, and it is dull enough that it stops happening around the second busy quarter. If you move it in-house, put it in a named person's calendar with a date, and keep the format identical to the supplier's so the series stays continuous.
Page updates move well when the pages are yours and badly when they are not. A content team that wrote the original article can usually add the missing answer to a section and tighten a heading. Rewriting a product page that marketing, sales and legal all have opinions about is where internal versions stall, which is worth being honest about before you take it on.
What tends to stay outside is the diagnosis and the off-site work. Reading a run log against a competitor's published pages, deciding which of your pages to fix first, and getting into the third-party lists the engines quote all need practice and time that most teams do not have spare. Our comparison of doing AEO in-house against an agency or a tool works through the staffing honestly, and whether you need third-party mentions covers the off-site half.
What are the signs you are paying for activity rather than movement?
Paying for activity rather than movement shows up in the shape of the invoice, not in the tone of the report. Four patterns are worth checking before a renewal, and each one has a cheap test.
Output drift is the first. Early in an engagement most of the gain comes from pages that already exist and nearly work, so the mix should be weighted towards rewriting. A report whose work log is almost entirely new articles in month nine is describing a content subscription, whatever the invoice says.
Unverified reporting is the second, and it matters more than it sounds. AirOps' 2026 analysis of the ChatGPT fan-out pipeline reports that only around 15% of retrieved pages end up cited, so retrieval, impressions and visibility scores can all move while citations do not. Ask which number in the report is a count of citations and which is a proxy.
Scope creep in the question set is the third. A set that grows every month makes the percentage look stable while the underlying position changes, because you are measuring a different thing each time. Freeze the set for a quarter at a time and version it.
The fourth is a report with no decisions in it. Every month should name what was changed, what is being tested next, and what would make the supplier change course. A report that only describes conditions is a weather forecast you are paying a retainer for. How many prompts you need to track explains why set size and set stability both matter to the number you are reading.
How do you run the renewal conversation?
Run the renewal conversation by asking for three priced options rather than a yes or a no, because the useful answer is usually in the middle. Suppliers who do this work seriously will have a maintenance tier whether or not it appears on their rate card.
Ask for the full scope, a maintenance scope and an exit plan, with what each includes written down. The maintenance quote tells you a great deal: a supplier who cannot describe a smaller version of the work is either selling a fixed product or does not know which parts of their own service produce the result.
Put exit criteria in the renewal rather than in the argument you will have later. Name the condition under which you would stop, such as being named in a stated share of answers across a frozen question set for two consecutive quarters, and agree what happens then. A review point every quarter with a stated right to stop costs a good supplier nothing.
Keep the market context in proportion. Conductor's January 2026 survey of more than 250 enterprise executives and digital decision makers found 94% planned to increase answer engine optimisation investment in 2026, which is a fact about enterprise budgets rather than an argument that you personally should spend more. Your own measured position is the better guide.
If the renewal rests on a number you cannot check, get an independent reading before the meeting. A free visibility assessment runs your buyer questions across ChatGPT, Claude, Perplexity and Google AI Overviews and reports who is named today, from which page, with the raw answers attached.
What else do people ask about when to stop paying for AEO?
How long should you give AEO before deciding to stop paying?
Give the work two full quarters before judging the outcome, and judge the measurement at the end of month one. Those are different tests. By week four you can tell whether the runs are real, repeatable and documented, which is a question about the supplier. Whether your position has moved is a question about the channel, and it needs enough repeated runs to separate a change from the ordinary rotation of sources between runs.
Will your AI citations disappear immediately if you cancel?
Citations do not disappear on the day you cancel, and they do not stay still either. Published decay studies disagree on the exact half-life because they define it differently, but all of them find the cited set rotating within weeks rather than months. A page that answers a question better than the alternatives can hold its place for a long time, while a page that was only marginally better than the next option tends to drop out first.
Is it cheaper to stop AEO and restart later?
Stopping and restarting costs more than a reduced scope, because restarting means rebuilding a baseline you already owned and re-reading pages somebody already read. The expensive part of a restart is not the setup fee. It is the lost comparability: a new series cannot tell you whether a page lost its citation during the pause or had already started drifting before it, so the first quarter back is spent re-establishing what you used to know.
What does a maintenance AEO scope usually include?
A maintenance scope is usually the measurement plus a small fixed amount of page work: the same question set run on schedule, a short report, and a couple of page updates a month where the log shows a loss. What it drops is new coverage, which is the part that scales with ambition rather than with risk. The test of a maintenance tier is whether it would still catch a competitor displacing you within a month.
Should you stop paying if your competitors stopped too?
Base the decision on your own measured position rather than on what competitors appear to be doing, because you cannot see their spend and you can see your own answers. If the engines name three rivals and not you, the gap is the argument for continuing regardless of their budgets. If you are named consistently across the question set and nobody is displacing you, that is a case for maintenance whatever anybody else is spending.