Answer engine optimisation is sold as a new discipline. Most of it is not. If you have done local SEO properly for a service business, the inputs an AI answer draws on are largely already in place: your Google Business Profile, your citations, your reviews and your pages.
The remaining part is real, though, and it is different enough that treating it as "SEO but for robots" produces bad advice. This post separates the two.
What is the same
Start here, because it saves money.
Google Business Profile still does the heavy lifting. Assistants read pages that quote it, and Google's own AI surfaces read it directly. Category, service list, hours, review count, review text. A neglected profile leaves both surfaces working from a thin record of what you do.
Citations and consistency still matter. Name, address and phone consistent across directories. Same reason as always: it is how any system, statistical or not, decides two records are the same business.
Reviews still matter, in volume and in content. The difference is that answers quote review themes rather than counting them. Otherwise the work is identical: ask, respond, do not fake.
Content that answers a real question still wins. A page that plainly states what you do, where, and for whom is as useful to a model composing a paragraph as it is to a person skimming.
If a vendor tells you AEO requires abandoning any of the above, that is the moment to stop the meeting.
What genuinely differs
The shortlist is a handful, not ten
A results page has ten organic slots plus a map pack. An assistant's answer has room for only a few business names before the paragraph becomes unreadable. Count the names in a few answers for your own category and you have the real number for your market.
This changes the economics of being nearly good enough. In search, position 8 gets a small share of clicks. In an AI answer, the equivalent of position 8 gets nothing at all, because it is not in the paragraph. There is no long tail of partial credit.
For a business, that means the distribution of outcomes is more brutal and more binary. For an agency, it means the pitch is easier: not "we can move you from 8 to 4" but "you are not in the answer, and here are the three companies that are".
The citation set is small and knowable
Ranking is decided by an enormous, opaque model of the whole web. An AI answer, in practice, is composed from a handful of pages the assistant fetched a second before it wrote.
You can see that set. Ask the question, ask for sources, and write down the URLs. Do that for all eight phrasings and you find out how much the list repeats; the pages that come back under several phrasings are the ones to start with. For ChatGPT, how to show up in ChatGPT recommendations shows where the sources appear and how to check that its search crawler can reach the site at all.
That is a materially different job from SEO. Instead of "improve authority", it is a finite checklist: get onto these specific pages, or get the facts on them corrected.
Being quoted beats being ranked
A model writing a recommendation needs a clause. "X is known for same-day emergency callouts" comes from somewhere. Businesses that state specific, checkable facts get those facts recycled into answers. Businesses whose whole online presence is adjectives get skipped, because there is nothing to compress.
The practical version: replace "professional, reliable, trusted" with "24 hour callout, family run since 2009, gas certified, fixed price quotes". Every one of those is a clause a model can lift.
Measurement is different, and worse
SEO has rank tracking. It is imperfect, but a rank is a rank.
AI answers are not deterministic. Ask twice, get different names. So the unit of measurement has to be a rate across many phrasings, not a position on one. That is why any honest audit asks the same question several ways and reports "named in 3 of 8" rather than "position 2".
It also means the tooling has to show its work. If a tool tells you a score with no way to read the underlying answers, you cannot check it and neither can your client. Our methodology page exists for exactly that reason: it describes the prompts, the matching threshold, and the cases where the method is approximate.
Attribution is currently bad
This is the honest weak point of the whole field. When someone asks an assistant for a plumber and then calls the plumber, there is often no referrer, no UTM, no click. The lead arrives as a phone call that looks like it came from nowhere.
You can partially close this: ask on the call, add a "how did you hear about us" field, watch for direct-traffic and branded-search lifts that track your visibility work. But anyone promising clean last-click attribution from AI answers today is selling something.
Say this out loud to clients before they ask. It is the difference between a hard question in month three and a hard conversation in month three.
What this means for the work you sell
If you run an agency, the practical shape of an AEO engagement looks like this.
Month zero, the audit. Eight phrasings, at least two assistants, logged out, fresh conversations. Record every name. Produce one table: the client at 0, and the businesses the answers name instead, with their scores. Collect the cited URLs. The 15-minute manual method is this audit written out step by step.
Month one, the boring wins. Google Business Profile completed properly, primary category corrected, services list written out. Listings on the specific directories that showed up in the citation set. Name consistency fixed across everything.
Month two, the quotable facts. One page per question shape you want to win, written in short declarative sentences. Specific differentiators replacing adjectives. Review requests aimed at the theme you want quoted.
Month three, the re-audit. Same eight questions, same conditions. Show the movement, including where there was none.
That is a real three month engagement with deliverables a client can see, and none of it requires believing anything unproven about how the models work.
Pricing it differently from SEO
Two things justify a different price from a standard local SEO retainer.
The audit is a discrete, valuable artefact. It has a before and an after and it takes real work to do properly. It can be sold on its own.
The scarcity is genuine. A handful of slots per question in a city is a much smaller prize than a page of ten results, so the value of holding one is higher, and the cost of ignoring it for a year is higher too.
What does not justify a different price is mystery. If your method is "we do AEO", you are charging for a word. If your method is "here are the eight questions, here is your score, here are the four pages that decide it, here is the plan", you are charging for work. The pricing guide for agencies turns that into three products with a deliverable each.
The uncomfortable summary
AEO is not a new discipline. It is local SEO with a smaller shortlist, a visible citation set, a preference for quotable facts, and worse attribution.
That is not a criticism. A smaller shortlist with a knowable citation set is the most tractable optimisation problem local search has had in years. When a business is at zero, the candidate explanations are unglamorous and cheap to check: the wrong primary category, absence from the pages the answers cite, a website made of adjectives.
Finding out which businesses are at zero, at volume, is the part that does not scale by hand. That is what AIgnored is for: it asks its own fixed set of 8 broad recommendation questions per model, scores every business in the city, and marks the ones with real reviews and a real website that the assistants never name. It does not ask the narrow service or price questions, so those stay a manual check. The pricing page has a free plan that covers a single city if you want to check its numbers against an audit you ran yourself.