The honest answer

Is AEO worth it?

The conditions under which it pays, the conditions under which it does not, and how to find out which you are in before spending anything.

AEO is worth paying for when buyers research your category through assistants before contacting anyone, one customer is worth several thousand, and competitors are already being named in answers where you are not. Outside those conditions it usually is not.

Every page answering this question argues from conviction. This one sets out the conditions, admits what cannot be proven yet, and gives you a test you can run yourself in ten minutes without buying anything.

The four conditions

All four have to hold. Three out of four usually means wait.

ConditionWhy it mattersFails when
Buyers research before contacting youAssistants are consulted during consideration, not at the point of impulsePurchases are immediate, local or price-driven
A customer is worth several thousandThe arithmetic has to survive a few months of no movementLow ticket, high volume — paid channels will beat it
Competitors are being cited and you are notThis is the measurable gap, and it is somebody else’s win recorded in writingNobody in your category is cited — the category question is not being asked yet
Your pages can actually be reached and quotedEverything else is irrelevant if the model cannot open the pageNothing — this one is fixable, and cheaply

The test, run it yourself

Open an assistant. Ask the question your best-fit buyer would ask — the category question, not your company name. “Best X for Y”, “how do I solve Z”. Then read what comes back.

  • Competitors named, you absent. That is the gap, and it is worth closing.
  • Nobody named, generic answer. The category is not being resolved to vendors yet. Early — cheap to prepare for, not urgent.
  • You are named. Good. The work becomes defending it, which is a different and smaller job.

Run it more than once. Assistant output varies between sessions, and a single run tells you less than it appears to.

What nobody can honestly promise you

No provider can guarantee a citation. Model outputs are non-deterministic, they vary by session, and they change with every model update. Anyone quoting a guaranteed position, a fixed percentage uplift or a projected revenue figure is describing something they cannot control. What can be promised is measurement on a fixed protocol, fixes to the things that demonstrably block citation, and re-measurement on the same basis.

The honest state of the evidence

This discipline is young enough that most published numbers are vendor-produced and most case studies are single-client anecdotes. We publish the third-party figures we rely on, name who produced each one, and state plainly on that page what they do not prove. The evidence page is here, including its own limitations section.

What we do not publish: client results we do not have, projected ROI, or percentage uplifts attributed to work whose causation we cannot isolate.

When it is not worth it — said plainly

  • Low-ticket or impulse purchases. Paid acquisition will out-earn it.
  • Local services where the buyer picks from a map, not an answer.
  • Companies with no content at all. There is nothing to make quotable yet; that is a content problem first.
  • Anyone needing attributable pipeline this quarter. Citation share moves over eight to twelve weeks, not eight days.
  • Categories where assistants return generic answers with no vendors named. Come back in six months.

The cheapest way to decide

Run the test above yourself. If it points at a gap, get the free check — it costs nothing, arrives in 48 hours, names who was cited instead of you, and gives you three fixes you can implement without hiring anyone. If the answer is that you are fine, we will say so.

FAQ

Related questions

The mechanical parts are verifiable: unblocking a retrieval agent measurably changes whether a model can fetch your page, and entity clarity changes whether it names you. What cannot be honestly claimed is a guaranteed citation or a predictable percentage uplift, because model outputs are non-deterministic and change with every model update.

Technical fixes apply immediately — unblocking an agent works the day it ships. Citation share moves more slowly: expect eight to twelve weeks before a fixed prompt set shifts meaningfully, because models have to re-encounter and re-trust the pages.

Low-ticket or impulse purchases, local services chosen from a map, companies with no content to make quotable yet, and anyone who needs attributable pipeline inside a quarter. In each of those cases the money does better elsewhere and we will say so.

Not cleanly, and anyone claiming otherwise is overstating it. Assistant referrals frequently arrive with the referrer stripped, and a buyer who first encountered you inside an answer may convert through a branded search weeks later. What can be measured is citation share on a fixed prompt set, AI referral traffic where it is detectable, and leads that self-report the source.

Ask an assistant the category question your buyer would ask — not your company name. If competitors are named and you are not, that is the gap. Run it several times, because output varies between sessions.

Settle it with data rather than opinion.

The free check names who is cited instead of you. 48 hours, no call, yours either way.