Answer engine optimization, or AEO, is the acronym currently being sold hardest, and the honest version of what it means will save you money.
Here is the first thing to know, and it reframes everything else. The term is not new and it did not come from generative AI. Trade press was writing about "the rise of Answer Engine Optimization" in February 2018, attached to voice search and featured snippets, five years before anyone was optimizing for a large language model. AEO is a 2018 snippet-era term retrofitted to 2026 chatbots.
Generative engine optimization, GEO, is the one with an academic paper behind it, published at a major conference in 2024. So the older, more marketed term is the one with the weaker provenance, which is the reverse of what most vendor blogs imply.
We have already covered the substance of this territory: what generative engine optimization actually is, why citations and rankings have decoupled, how to write for citation, and how to measure it by hand. This piece is deliberately not another one of those. It is the terminology and evidence audit: which of the five acronyms you can safely ignore, which tactics have controlled evidence behind them, which have controlled evidence against them, and what the measurement layer actually costs.
The short version: AEO and GEO are the same thing under two labels, the discipline is SEO with its center of gravity moved off your own site, and the genuinely new thing is a measurement industry with no ground truth, priced from $29 to six figures a year.
Nobody can define the difference, including the people selling it
Five sources, in descending order of authority.
Google says it is SEO. Its official guidance on AI optimization, updated July 2026, names AEO and GEO explicitly and says:
"From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO."
Forrester says it is not fundamentally different. Analyst Nikhil Lai: "AEO is significantly, but not fundamentally, different from SEO. The practices are aligned but have technical differences."
The academic literature has no consensus definition. As of early 2026, no distinction between AEO, GEO, LLMO and AI SEO had been established, and the terms are used interchangeably in practice.
The biggest AEO vendor says the terms are identical. Profound, one of the largest pure-play measurement companies, published a post in June 2025 titled "AEO vs. GEO: Why they’re the same thing (and why we prefer AEO)." Its stated reason for preferring AEO is that GEO "is not ownable." That is a naming-rights argument, not a technical one.
And the analyst category that exists is for tools, not for a discipline. Gartner published a Market Guide in March 2026, and its title is "Answer Engine Visibility Tools." Several vendors, including at least one accessibility and content platform, have described it in their own press releases as a market guide for answer engine optimization. That slippage is worth noticing, because it is the whole business model in one word.
What answer engine optimization actually changed
Two things are genuinely new. Neither is a new discipline, and both are worth your attention.
The unit of success moved from a ranked URL to a named entity. SEO optimizes a document to rank in a list. An answer engine can cite your page without naming your brand, or name your brand without citing you. Seer Interactive quantified the gap across 541,213 responses from six AI platforms in February 2026: when a brand is mentioned, its citation rate is 53.1 percent; when it is not mentioned, 10.6 percent. Their summary is the useful one: "Your content gets cited. Your competitors get recommended. You don’t."
That is a failure mode page-level SEO has no vocabulary for, and it is real.
The leverage moved off your own site. Semrush analyzed 126 million US AI prompts between January and April 2026 across ChatGPT, Gemini, Google AI Mode and AI Overviews, and found visibility is built through "customer reviews, community platforms, independent publishers, and retailers, not solely owned websites." It also found that being mentioned in an answer does not mean your own site is cited as the source.
That is a budget reallocation, not a new technical skill. It moves money from on-page work toward digital PR, community presence, trade press and reference sources. For an association or nonprofit, that means directory accuracy, Wikipedia and Wikidata correctness, trade publication mentions and genuine community participation.
What did not change: crawlability, server-side rendering, clear headings, factual accuracy, freshness, topical depth and entity consistency. Every credible source lands on the same foundation.
There is a fair counter-argument worth hearing. Mike King of iPullRank argues the "it’s just SEO" framing is self-serving and hides real work, and his strongest technical point is this: "chunking is what RAG systems do to your content, whether you optimize for it or not. The question is whether your content survives the chunking process with its meaning intact." Retrieval-aware structuring is a genuine, if modest, technical addition. It is not a discipline.
The two tactics with controlled evidence against them
This is the section that will save you the most money.
Schema markup does not measurably increase AI citations. Ahrefs ran a matched difference-in-differences study in May 2026: 1,885 treated pages against 4,000 matched controls, treatment window August 2025 to March 2026, thirty days before and after.
The correlational finding first, because it is the one vendors quote: cited pages were almost three times more likely to have JSON-LD than non-cited pages.
The causal finding: Google AI Overviews minus 4.6 percent, AI Mode plus 2.4 percent, ChatGPT plus 2.2 percent. The last two are indistinguishable from zero. Their conclusion was that adding JSON-LD did not boost citations on AI Mode or ChatGPT and showed only a small decline in AI Overviews that could not clearly be attributed to schema.
Google agrees: "Structured data isn’t required for generative AI search, and there’s no special schema.org markup you need to add."
Keep the schema you have for rich results. Stop paying anyone to add it for AI citations. That is the correlation-versus-causation split your clients most need explained, and a correlational study claiming a 22 percent lift is measuring the same thing the controlled study found no effect from.
llms.txt is essentially unread. Ahrefs analyzed 137,210 domains in June 2026 using server logs. Twenty-eight percent published an llms.txt file. Ninety-seven percent of those files received zero requests in the month studied. Of the three percent that got any traffic, most was SEO audit tools and llms.txt-checking tools; actual AI retrieval bots were around 1.1 percent combined. No AI bot probed for the file on domains that did not have one.
Google’s position: you do not need to create AI text files to appear in Search, and doing so "will neither harm nor help your site’s visibility or rankings." Its June 2026 update softened that to permission rather than endorsement.
No major AI lab has publicly stated that its production systems read or act on llms.txt.
Publishing one costs ten minutes and does nothing measurable. Paying anyone for llms.txt optimization is not defensible.
What does have evidence behind it
Off-site brand mentions are the strongest measured correlate, by a distance. Ahrefs analyzed 75,000 brands and reported Spearman correlations with AI Overview visibility:
| Factor | Correlation |
|---|---|
| Branded web mentions | 0.664 |
| Branded anchors | 0.527 |
| Branded search volume | 0.392 |
| Domain Rating | 0.326 |
| Referring domains | 0.295 |
| Backlinks | 0.218 |
The authors caveat it themselves: correlation is not causation. But note the shape. Brand mentions outrank backlinks three to one. That is the single finding that should change a budget.
Statistics, citations and quotations are the only tactics with a peer-reviewed effect size. The GEO paper tested around 10,000 queries and found that adding statistics, citations and quotations produced the largest visibility gains. It also found keyword stuffing sometimes performed worse than baseline.
Serve rendered HTML. Analysis of AI crawler behavior has reported no evidence that these crawlers execute JavaScript, which makes server-side rendering a real requirement rather than a nice-to-have. Worth verifying against the primary source before you rebuild anything, but it is the highest-certainty technical item on the list.
Everything else that circulates, the freshness percentages, the YouTube share figures, the page-speed thresholds, the E-E-A-T lift numbers, comes from vendor blogs with varying methodological disclosure. Treat those as directional.
The measurement layer, which is the actual product being sold
Here is what the tools cost, and what they do.
| Tool | Published price |
|---|---|
| Otterly.AI | from $29/mo |
| HubSpot AEO | $45 to $50/mo |
| Semrush AI Visibility Toolkit | $99/mo per domain |
| Profound Starter | $99/mo, ChatGPT only |
| Profound Growth | $399/mo |
| Profound Enterprise | reportedly $30,000 to $100,000+/yr |
| Ahrefs Brand Radar | from $199/mo |
| Peec AI | around €75/mo |
| Siteimprove AEO Insights | not published |
The methodology is uniform and simple. They run a fixed prompt set against the engines on a schedule and count brand mentions and cited domains. That is it. It is a synthetic panel, not a measurement of real user queries.
Four structural limitations, all documented, and together they are damning:
The same engine gives different answers to the same prompt. Semrush ran 100 prompts twice in June 2026, once at minimal reasoning and once at high, across 20 buyer journeys. Only 25.6 percent of cited domains overlapped. Nearly three in four cited sources were different. Any tool treating "ChatGPT" as one system is measuring noise.
Engines barely overlap with each other. Roughly 11 percent domain overlap between ChatGPT and Perplexity in one large study. A blended "visibility score" is not aggregatable.
The prompt set is chosen by the vendor or by you, so the denominator is arbitrary. There is no equivalent of Search Console impressions.
Trackers disagree with each other on the same brand.
Is there any ground truth? Partially, and only on one side. Google Search Console now has a generative AI report, which is first-party and authoritative for Google surfaces. Your server logs give you verifiable AI crawler fetches. Referral analytics give you verified arrivals, though attribution is badly broken: one analysis of 446,405 visits found 70.6 percent of AI traffic misattributed as direct in GA4.
Everything else is a vendor’s synthetic panel and should be described that way when you present it.
The traffic picture, honestly
Volume is down and still falling. Quality per visit is up several-fold. Absolute AI referral volume is still tiny.
The best-designed public study remains Pew’s, published July 2025: a real browsing panel of 900 US adults, 68,879 searches, 12,593 producing an AI summary. Users clicked a result on 8 percent of visits with an AI summary against 15 percent without. Clicks on citations inside the summary: 1 percent.
Zero-click is accelerating. SparkToro’s analysis of US clickstream data for January to April 2026 found 68 percent of Google searches ended without a click, up from 60 percent in 2024 and 49 percent in 2019.
Take the counter-evidence seriously. Graphite analyzed the 40,000 largest US websites from February 2024 to November 2025 and found SEO traffic down only 2.5 percent, not catastrophically. Their mechanism: AI Overviews cut click-through by around 35 percent when present, but appear only about 30 percent of the time. And they disclosed their own conflict, noting it is in the author’s financial interest to argue SEO is not declining. That disclosure is rarer than it should be and earns the finding a hearing.
But publishers specifically are hit far harder than the web average, with reported declines around a third against Graphite’s 2.5 percent across all sites. For a publisher audience, the publisher number is the relevant one.
On quality, Adobe reported AI-referred conversion at 1.66 percent against 0.15 percent for search in early 2026, roughly eleven times better. And Ahrefs found ChatGPT handles around 12 percent of Google’s search volume while sending 190 times less traffic.
The honest formulation: a ten-times-better conversion rate on one percent of your traffic does not replace seven percentage points of lost click-through. Say that to a client before they reallocate a budget.
What to actually do
Do these, all cheap and all evidence-backed:
Serve rendered HTML. AI crawlers show no evidence of executing JavaScript.
Answer the question in the first hundred words, then support it. Your content gets chunked whether you plan for it or not.
Include original statistics, named sources and direct quotations. The only tactic with a peer-reviewed effect size.
Spend on being mentioned elsewhere rather than on markup. Brand mentions correlate three times more strongly than backlinks.
Keep entity data consistent across your site, Wikipedia, Wikidata and directories.
Measure with what is verifiable: Search Console’s generative AI report, server logs for crawler fetches, and referral data with the attribution caveat understood.
Publish an llms.txt if you like. Ten minutes, no measurable effect, no harm.
Do not do these:
Add schema for the purpose of AI citations. A controlled test found no effect and Google says it is not required.
Pay anyone for llms.txt optimization.
Buy a $400 a month tracker for a site with no AI visibility to track. Start with Search Console and logs. A tracker earns its cost once you have a prompt set tied to revenue.
Treat an AI visibility score as a KPI. A 25.6 percent source overlap between two modes of the same model makes it unstable by construction.
Deliver different content to bots. That is cloaking.
Rebuild your content strategy around a new acronym.
One budget shape worth borrowing, from EMARKETER: 40 percent core SEO, 25 percent digital PR, 20 percent data and reporting, 10 percent training, 5 percent experimentation.
Frequently Asked Questions
What is answer engine optimization?
Optimizing content to be surfaced and cited by AI answer engines rather than only ranked in a list of links. The term dates to 2018 and voice search, and was retrofitted to large language models around 2024.
Is AEO different from GEO?
In practice, no. The terms are used interchangeably, no consensus academic definition separates them, and the largest pure-play AEO vendor published a post arguing they are the same thing.
Is AEO different from SEO?
Google says optimizing for generative AI search “is optimizing for the search experience, and thus still SEO.” Forrester calls it significantly but not fundamentally different. The genuine changes are that success is measured on entity mentions rather than ranked URLs, and that leverage has moved off your own site.
Does schema markup help AI citations?
A controlled study of 1,885 pages against 4,000 matched controls found no significant uplift on AI Mode or ChatGPT. Google says structured data is not required for generative AI search. Keep schema for rich results, not for citations.
Does llms.txt work?
A study of 137,210 domains found 97 percent of llms.txt files received zero requests in the month measured, and actual AI retrieval bots accounted for around 1.1 percent of the traffic the remainder got. No major AI lab has said its production systems use it.
What actually influences AI citations?
Off-site brand mentions correlate most strongly, at roughly three times the strength of backlinks. Adding statistics, citations and quotations is the only tactic with a peer-reviewed effect size behind it.
How do I measure AI visibility?
Search Console’s generative AI report and your own server logs are the only verifiable sources. Commercial trackers run fixed prompt sets on a schedule, which is a synthetic panel rather than a measurement of real queries.
Are the tracking tools worth buying?
Not until you have a prompt set tied to revenue. They range from $29 a month to six figures a year, and their outputs are unstable enough that one model gives 25.6 percent source overlap between two of its own reasoning modes.
Is AI traffic replacing search traffic?
No. Zero-click Google searches reached 68 percent in early 2026 and AI tools collectively send under one percent of all traffic. Conversion quality on that traffic is several times better, which does not make up the difference in volume.
Should I hire an AEO agency?
Ask what they would do differently from good SEO plus digital PR. If the answer centers on schema packages or llms.txt optimization, the controlled evidence is against both.
Do I need a separate AEO strategy?
No. You need the SEO fundamentals done properly, plus a deliberate shift of budget toward being mentioned on sites you do not own, plus measurement using the sources that can actually be verified.
What is the one thing worth changing today?
Move money from on-page markup to off-site presence. That is the only reallocation the correlational and controlled evidence both point toward.
