A new discipline is being sold to marketing teams right now, with new acronyms (AEO, GEO, answer engine optimization), new retainers, and a familiar promise: the rules have changed, and you need a specialist to survive them.
Our argument, having read the evidence: most of what is sold as answer engine optimization is editorial hygiene. The discipline is real. The mystique is not. And the genuinely new parts are not the parts the retainers emphasize.
The pitch, and why it lands
The fear underneath the pitch is legitimate. Half of B2B buyers now begin software research with an AI chatbot more often than with Google, per G2's April 2026 study of 1,076 decision-makers.
Into that anxiety, an industry has arrived selling answer-engine visibility as proprietary knowledge. A black box invites a priesthood.
Disclosure: this publication is produced with support from Docket, which sells software in this category. The argument below mostly cuts toward doing the work yourself; weigh it knowing where we sit.
What the one controlled study found
The closest thing this field has to a controlled experiment is GEO: Generative Engine Optimization, from researchers at Princeton and collaborating institutions, presented at KDD 2024. The team built GEO-bench, 10,000 queries across domains, and tested a battery of content changes for their effect on how often, and how prominently, a page shows up in generative answers.
Content changes alone moved visibility by up to 40%. Look at which changes did the moving: adding citations to sources. Adding quotations. Adding statistics. Even pure fluency edits, clearer sentences with no new information, produced double-digit gains in the paper's results.
That list is not a new discipline. It is the checklist a demanding editor has enforced since before the web: support your claims, quote people who know, use real numbers, write so you can be excerpted.
What the engine's own documentation says
Google publishes guidance for site owners on AI features, and it is striking mostly for how boring it is: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.” And: “You don't need to create new machine readable files, AI text files, or markup to appear in these features.” Eligibility is the old floor, indexed and eligible to show a snippet, and the recommended practice is the old practice: helpful, reliable, people-first content.
Platform owners can be wrong about their own platforms, and every party here has incentives. But when the seller of the mystique is contradicted by the operator of the engine, the burden of proof sits with the mystique.
The answer-hygiene checklist
Here is what we would do to a page you want cited. Most of it is an afternoon of editing per page.
Lead with the answer. The first two sentences under a heading should answer the question the heading asks. Context comes after.
Make headings the questions buyers actually ask. One question per section. If a section answers two questions, split it.
Cite a primary source for every factual claim. Link the study, not a blog post about the study.
Use real numbers where they exist. Statistics were among the strongest levers in the Princeton results. “Significantly faster” is not a number.
Quote someone accountable. A named person saying a concrete thing beats a stock phrase from a brand voice.
Edit for extraction. Short sentences, one idea each. If a paragraph cannot survive being lifted out alone, rewrite it.
Check the mechanical floor. Indexed, snippet-eligible, and not blocking the crawlers you want citing you (OpenAI's, Perplexity's, Anthropic's) in robots.txt. And the answer cannot live behind a form. A gated answer is invisible to every engine, and to most buyers.
Where hygiene stops
Two problems on this list are genuinely new, and neither is on-page work.
Measurement. Assistants pass along little clean referral data, and the tools that track AI citations are young and frequently disagree. Our view: imperfect tracking of a real trend beats precise tracking of a fading one. But budget for ambiguity, and treat every AI-visibility dashboard as directional.
Authority. Hygiene makes a page easy to cite. It does not make your company the answer. Generative engines synthesize from the public record, and if that record barely mentions you, the fix is earned coverage, original research, and third-party corroboration, work that lives in PR and product marketing, not in markup. This is where we believe the serious version of the discipline earns its fee.
One marketing operations leader we spoke with this month had started buying Reddit ads for a reason that had nothing to do with Reddit's audience. They had watched their own brand begin surfacing in ChatGPT answers, believed Reddit threads were part of what the models had ingested, and were buying presence there as a route into the answer rather than a route to a click. Neither they nor we can verify that mechanism, and they said so themselves. We include it because it is the clearest signal we have that authority work is now being aimed at a reader that is not a person.
A one-quarter operating cadence
Write down the twenty questions your buyers actually ask. The people taking sales calls hear them every day. Once a month, put each question to ChatGPT, Perplexity, and Google's AI Overviews. Record whether you are cited, who is cited instead, and what their page does that yours does not.
Then run the checklist on the ten pages that should answer your money questions, and re-test the next month. Report the share of buyer questions where you appear. Not rank: share of answers. That is a number a demand generation team can act on, and it survives contact with a skeptical CFO.
FAQ
Is AEO different from SEO? The floor is identical: indexed, crawlable, snippet-eligible, genuinely helpful content. The emphasis differs: answer-shaped pages, cited claims, real numbers, and extractable writing matter more when the reader is a model assembling an answer rather than a person scanning links.
Do we need special AI files or markup? For Google's AI features, Google's documentation says no. Other engines publish less guidance, but the strongest measured effects in the public research came from content changes, not markup.
Can we measure AI search traffic? Partially. Referral data from assistants is thin, and citation trackers disagree with each other. Track the share of buyer questions where you appear, treat it as directional, and resist precision theater. Apply the same scepticism to the adoption statistics quoted at you, which usually measure less than they appear to.
Should we hire an AEO agency? Ask what they would change on your ten most important pages. If the answer is hygiene, measurement, and earned authority, they are selling the real thing. If it is a proprietary methodology they cannot explain, you are buying the priesthood.