Answer engine optimization is the practice of making a source easy for AI systems to retrieve, extract from, and confidently attribute. The 2026 evidence points to five things that matter: machine access, extractable structure, corroboration from other sites, unambiguous entity identity, and measurement. Structure and off-site corroboration carry more weight than most on-page tactics being sold as AEO.
I have read a great deal of AEO advice over the past year, and the thing that strikes me is how little of it is checkable. The format is nearly always the same. A confident list of tactics, no source, no measurement, and an implication that following them produces citations.
Some of it is probably right. The problem is that you cannot tell which parts, and neither can the people writing it.
So this is an attempt at the opposite. Everything below is either backed by published 2026 research, which I have cited so you can go and disagree with it, or flagged as uncertain. There is more in the second category than the industry generally admits.
What AEO actually means
Answer engine optimization is the work of becoming a source that AI systems retrieve, quote, and attribute when they generate an answer.
That is a different objective from ranking. A ranking contest is about position in a list of links. A citation contest is about whether a model can find your claim, lift it cleanly, and feel confident enough to name you as the source.
How it relates to SEO and GEO
These terms are used loosely and it causes real confusion, so here is the distinction I use.
SEO earns position in a results page. AEO earns inclusion in a generated answer, whether that is an AI Overview, ChatGPT, Perplexity, or Gemini. GEO, generative engine optimization, is used interchangeably with AEO by most practitioners, though some reserve it for generative systems specifically.
The practical point is that they share most of their foundations. A page that search engines cannot crawl will not be cited either. Treating AEO as a replacement for SEO is the most expensive mistake available in this area.
The finding that reframes everything
Before the tactics, one piece of research deserves to sit at the top, because if it holds up it changes the priority order completely.
Seer Interactive's 2026 analysis suggests LLM citation may be substantially post-hoc. That is, the model appears to decide which brands to recommend from what it already knows, then retrieves sources to support that choice, rather than researching neutrally and citing what it finds.
Why that matters more than any on-page tactic
If citation is post-hoc, then the decisive factor is whether the model already knows you exist. That is a product of your presence across the wider web: mentions, coverage, directories, forums, other people's content. Not your heading structure.
I want to be careful here. This is one firm's analysis, not settled science, and it may be partially true rather than wholly true. But it is directionally consistent with something practitioners keep observing, which is that well-optimized pages from unknown brands underperform mediocre pages from known ones.
The working conclusion I would draw: on-page work is necessary and insufficient. Do it, and do not expect it alone to be enough.
The AEO Visibility Stack
Here is the model I use to structure the work. Five layers, in dependency order, because each one is wasted if the layer beneath it is missing.
| Layer | Question it answers | Failure symptom |
|---|---|---|
| 1. Retrievability | Can AI systems access the content at all? | Never cited, never appears in any tool |
| 2. Extractability | Can a model lift a clean answer from the page? | Page is crawled but competitors get quoted |
| 3. Corroboration | Does anything off-site confirm the claim? | Cited occasionally, inconsistently |
| 4. Entity clarity | Does the model know who you are? | Cited without attribution, or attributed wrongly |
| 5. Measurement | Can you tell whether any of it worked? | No idea whether to continue |
Work upward. Most sites I look at have problems at layer 2 and layer 3 while spending their effort on layer 1, which was already fine.
Layer 1: retrievability
The baseline. AI crawlers need to reach and render the content.
The most common failure I encounter is content that only exists after JavaScript runs. If your key text or your internal links are injected client-side, you are relying on the crawler to execute scripts, and many do not. This is the same failure that hurts traditional indexing, which is a recurring theme: AEO problems are frequently just SEO problems with a newer name.
Blocking is the other half. Several major AI crawlers respect robots.txt, so it is worth knowing whether yours is turning them away. That decision, and the llms.txt question, gets its own treatment in the technical cluster of this series.
Layer 2: extractability
This is where the published evidence is strongest, and where most sites lose.
Content that is explicitly structured, meaning tables, lists, and clear FAQ blocks, is cited roughly two to three times more often than prose-only content. Heading density matters too: spacing of about 120 to 180 words between headings correlates with roughly 70% more citations than sparse or inconsistent structure.
Position within the page matters as much as structure. Around 44.2% of LLM citations come from the first 30% of a page. Whatever your most quotable claim is, burying it under 800 words of context makes it substantially less likely to be used.
The single strongest configuration
The most citation-effective pattern identified in 2026 research combines two things: a direct answer capsule near the top, and proprietary insight the model cannot get elsewhere. That combination shows a citation rate of about 34.3%, materially higher than either element alone.
That is worth sitting with, because it says something uncomfortable about the average blog post. Restating consensus in a well-structured way gets you partway. Having something of your own to say is what closes it.
Format matters too, though less than people hope. Ranked "best-of" listicles are the single most-cited content format at roughly 21% of citations, per arXiv 2606.20065. Useful to know, easy to over-apply.
Layer 3: corroboration
A claim that appears only on your own site is a claim with one source. Models appear to weight agreement across independent sources heavily, which is sensible behaviour for a system trying not to be wrong.
This is the layer that the post-hoc finding elevates. Mentions on other sites, presence in directories and industry listings, being quoted by someone else, showing up in the places where your topic gets discussed. None of it is glamorous and none of it happens on your website.
For a small firm, the realistic version is narrow and deep rather than broad. Be the source that a specific niche keeps referencing, rather than a minor voice across a wide one.
Layer 4: entity clarity
Models need to resolve who you are as an entity, not just parse your page. That means consistent naming, a consistent description of what you do, and machine-readable identity via schema, plus sameAs links tying your site to your profiles elsewhere.
Inconsistency is the enemy. If your site describes you one way, your LinkedIn another, and your directory listings a third, you are making the resolution harder for no benefit. This is cheap to fix and frequently neglected.
Layer 5: measurement
You cannot manage what you cannot see, and AI visibility is genuinely harder to see than rankings. Answers are personalized, non-deterministic, and vary between sessions, which means point-in-time checks are close to useless.
The honest approach is directional: track mention frequency across a fixed prompt set over time, watch referral traffic from AI sources, and accept a wide error bar. The measurement cluster in this series covers how to build that without deceiving yourself.
The strategic advantage small sites actually have
One finding in the 2026 research deserves more attention than it gets, because it inverts the usual assumption.
Topical authority appears to beat domain size for citation purposes. A small site that covers one subject exclusively can outcompete a large general publication for citations on that subject, apparently because models treat focused sources as more reliable on their specialty.
In classic SEO, a small site fighting a large publisher usually loses on domain authority alone. In citation contests, depth in a narrow lane is a genuine competitive asset. That is the best structural news small businesses have had from search in some years, and it argues strongly for covering one thing thoroughly rather than many things adequately.
What does not appear to work
Worth being direct, because these consume real effort.
llms.txt, on current evidence. Adoption sits around 10% after eighteen months of discussion. Crawler interest has been measured at 408 llms.txt requests against more than 500 million AI bot visits over a 90-day window, and studies show no measurable citation lift from the file alone. Google's Gary Illyes stated in July 2025 that Google does not support it and has no plans to. It is also widely misunderstood: it is a routing file, not access control, and it does not opt you out of training. I would not tell a client it is useless, since it is cheap and the situation may change. I would tell them not to expect anything from it.
Keyword stuffing for AI. The systems are language models. Repetition does not persuade them.
Volume without depth. Publishing more thin pages appears to help neither ranking nor citation, and now competes with AI Overviews for the informational queries thin content usually targets.
Where I would start
If you are beginning from nothing, the order that respects the dependency stack is roughly this.
Confirm your content is reachable and rendered without JavaScript. Then restructure your best existing pages: answer capsule near the top, headings every 120 to 180 words, one table or list where it genuinely helps, an explicit FAQ. Then fix entity consistency across your site and profiles, which is a day of work. Then start the slow off-site work, which is the part that takes quarters rather than weeks. Then measure, directionally, and expect noise.
Notice that four of those five steps are things a competent SEO practitioner was already doing. That is not a disappointment, it is the actual finding.
The honest limits
AEO is roughly two years old as a discipline. Much of what is published, including some of what is cited above, rests on single studies that have not been independently replicated. Platform behaviour changes without notice and without documentation.
Anyone telling you they have a reliable system for getting cited is describing something nobody currently has. What exists is a set of practices that are individually well-founded, collectively unproven as a package, and mostly identical to doing genuinely good work that happens to be well-structured.
I would rather say that plainly than sell certainty that does not exist.
What this looks like for an actual small business
Frameworks are easy to nod along to and hard to act on, so here is the stack applied to a plausible situation.
Take a twelve-person accounting firm serving dental practices in a US metro. They have a website with a homepage, an about page, four service pages, and roughly thirty blog posts published over five years. They rank reasonably for their firm name and almost nothing else. Nobody has ever asked an assistant about them, and if anyone did, the assistant would not mention them.
Layer one is probably fine. The site is old, it is plain HTML or a standard content management system, it renders without JavaScript, and it is indexed. Ten minutes of checking confirms it, and this is the layer most people spend money on because it is the easiest to sell.
Layer two is where the loss is. Those thirty blog posts almost certainly open with three paragraphs of context before saying anything, run eight hundred words under a single heading, contain no tables, and end with a conclusion holding the only quotable sentence in the piece. Restructuring the best six of them is about a day of work and it is the highest-return day available to this firm.
Layer three is close to zero, and this is the real constraint. The firm exists in the state CPA society directory and nowhere else. No dental industry publication has ever mentioned them, they do not participate in the forums where dental practice owners discuss running their businesses, and no comparison article lists them. A model has no reason to know they exist, so no amount of layer two work will produce citations on its own.
Layer four is a day of work they have never done. Organization schema, consistent naming, sameAs linking to their LinkedIn and their society profile.
Layer five does not exist, so they have no way of knowing whether anything they do next helps.
The correct sequence for this firm is not what most proposals would suggest. It is: confirm layer one in an afternoon, spend a day on layer four because it is trivially cheap, spend a week on layer two across their best pages, set up layer five so the next twelve months are measurable, and then commit to layer three as a permanent activity, starting with the dental-industry venues where their actual buyers already congregate.
Notice what dominates the calendar. The first four items take about two weeks. The fifth takes years and is the one that decides the outcome. That ratio is the honest shape of AEO for a small business, and it is why so much of the advice concentrates on the two weeks.
Frequently asked questions
Is AEO replacing SEO?
No. AEO depends on the same foundations: crawlable content, clear structure, credible sources. What has changed is that a portion of search demand now resolves without a click, so visibility and traffic have partly decoupled. Treat AEO as an addition to the measurement and structure of your search work, not a replacement for it.
How long does AEO take to show results?
Layer 2 changes, meaning structure and answer capsules, can show up in weeks because they affect pages that are already retrievable. Layers 3 and 4, corroboration and entity presence, take quarters. If citation really is partly post-hoc, off-site presence is the slowest and most durable part of the work.
Do I need special AEO tools?
Not to begin. A fixed set of prompts run monthly against the main assistants, recorded in a spreadsheet, will tell you more than most paid tools at this stage. Buy tooling when the manual version becomes the bottleneck.
Does schema markup make me get cited?
It helps machines resolve what your page is and who published it, which supports layers 2 and 4. It is not a citation switch. Sites with excellent schema and no authority do not get cited, and sites with authority and no schema often do.
Should small businesses bother with AEO at all?
Yes, and the topical authority finding is the reason. Narrow, deep coverage of one subject is one of the few areas where a small site holds a structural advantage over a large one. That advantage does not exist in most other parts of search.
The uncomfortable summary is that the best-supported AEO advice looks a lot like doing genuinely useful work, structuring it so a machine can read it, and being known by more than one website. The tactics that feel novel are the ones with the least evidence behind them.
If you want a straight read on where your site sits in the five layers, email adphconsulting@gmail.com. ADPH is a US-facing consultancy that delivers its work from the Philippines, and I am happy to tell you when the answer is that your foundations need attention before any of this is worth doing.