AI SEARCH & AEO · September 2026 · ~10 min read
When AEO is just SEO with a new name
Most of the time. Google says optimizing for its generative AI search is optimizing for the search experience, and thus still SEO. The genuinely new parts are narrow: how you format an answer, how you measure a system that varies run to run, how you read crawler logs. Everything else on an AEO invoice is renamed SEO.
On this page
- 01What does Google actually say about the category?
- 02What is genuinely new?
- 03What is being renamed?
- 04What does the renaming actually cost you?
- 05Why does the renaming work so well?
- 06Does it matter if it is renamed, as long as the work happens?
- 07What to do this week
- 08When you do not need this
- 09Sources
- 10Related reading
- 11Got a proposal with AEO on it?
Renaming is not automatically fraud. Categories get new names all the time, and sometimes the name helps people understand what changed.
It becomes a problem in one specific case: when the new name is used to sell you work you are already paying for, at a premium, as though it were additional. That happens often enough that it is worth being able to spot in a document.
01What does Google actually say about the category?
Two sentences, and both of them are unusually direct for a company that normally hedges.
The first is the framing sentence, from Google Search Central's guidance on optimizing for generative AI features, published in May 2026 and updated in July: "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO."
The second is the instruction that follows from it. Google tells site owners to "prioritize effective SEO strategies over 'AEO/GEO hacks.'" The company that operates the surface you are being sold access to put the acronyms in quotation marks in its own documentation.
The same page states the technical requirement in full. A page must be indexed and eligible to be shown with a snippet. There are no additional technical requirements. Not a file, not a markup type, not a format.
Google then adds a warning aimed squarely at the tooling layer: "Be wary of third-party tools that promise ranking success or claim to use 'internal' Google metrics. No third-party tool has access to our internal ranking or AI systems."
None of this makes Google a neutral narrator. It has commercial reasons to keep site owners inside its existing playbook. But its negative claims are the ones that hold up against outside data, and they are the ones a vendor's invoice usually contradicts.
02What is genuinely new?
Three things, and they are smaller than the marketing suggests.
Answer formatting. Answering the question in the first three sentences, keeping paragraphs self-contained so they survive being lifted out, stating a definition plainly once. This was always good writing. It now has a mechanical payoff, because an assembled answer needs quotable units.
There is one peer-reviewed experiment behind that claim and it is worth knowing its actual shape. The KDD 2024 paper "GEO: Generative Engine Optimization" tested nine content tactics against a benchmark of 10,000 queries, scored on a position-adjusted word count metric. The do-nothing baseline scored 19.3. Adding direct quotations scored 27.2, adding verifiable statistics 25.2, adding inline citations 24.6. Keyword stuffing scored 17.7, below the baseline, meaning it actively hurt. Writing in a more authoritative tone produced no significant improvement at all, which the authors read as evidence that these systems already shrug off changes of that kind.
Two limits travel with that finding. The same tactic ranged from 19.1% to 35.8% in effectiveness depending on the subject category, so there is no universal number. And the experiment fed cleaned text from five fixed Google results to an older model, with nothing local or commercial in the benchmark.
Measurement. This is the real change. These systems are non-deterministic, so there is no position to track. A University of St. Gallen team ran four engines daily for six weeks and also ran identical prompts repeatedly within a single day. Between two consecutive days, only 34% to 42% of cited sources overlapped. Same day, same prompt, ten runs: 32% to 43%. Honest measurement means running a fixed prompt set repeatedly and reporting rates with a confidence interval. That method did not exist in the rank-tracking era and it is not a renaming of anything.
Crawler log analysis. Reading server logs to see which AI crawlers fetched which pages, verified by IP range rather than user agent string. Log analysis is old. Doing it for this purpose, and knowing the operators publish IP ranges for verification, is new work.
That is the honest list. Three items, none of which requires a separate department.
03What is being renamed?
Most of the rest.
Making your site crawlable and indexed is technical SEO. It is now sold as AI readiness.
Writing pages that answer real customer questions is content strategy. It is now sold as answer engine optimization.
Getting mentioned by relevant third-party sites is digital PR and link building. It is now sold as brand entity building.
Keeping your business details accurate across the web is citation and profile management. It is now sold as entity optimization.
Getting more reviews is review management. It is now sold as AI trust signals, though reviews genuinely do feed AI answers, so the underlying claim is sound and only the pricing is opportunistic.
Then there is the layer that is not renamed SEO but is simply nothing. Google names four items directly: special files such as llms.txt, artificial content chunking, rewriting content for AI systems, and seeking inauthentic mentions. Those four appear on invoices constantly.
The evidence on the first one is not just Google's word. Ahrefs examined every domain in its Web Analytics product with traffic in May 2026, 137,210 sites, and found that 97% of published llms.txt files received zero requests in that month from any bot or human. Of the requests that did happen, AI retrieval bots were 1.1%. Ahrefs sells SEO software, and its own caveat is that fetched does not mean read, so every figure it published is a ceiling rather than a floor.
If a line item is renamed SEO, question the price. If it is on Google's ineffective list, question the vendor.
04What does the renaming actually cost you?
Run the arithmetic on your own proposal. Here is the shape of it.
Take an AEO retainer quoted at $2,400 a month, on top of an existing SEO retainer of $1,800 a month. Twelve line items on the AEO scope. Sort them into three columns.
- Renamed SEO: 8 items. Technical crawlability, content production, citation cleanup, review prompts, internal linking, page speed, schema, reporting.
- Genuinely new: 2 items. A repeated-run prompt measurement program, and monthly crawler log analysis verified by IP.
- Named ineffective by Google: 2 items. An llms.txt build and an "AI content restructuring" pass.
Now price the columns. If the twelve items are billed evenly, each is $200 a month. The renamed column is $1,600 a month, or $19,200 a year, and you are already paying for most of it inside the $1,800 SEO retainer. The ineffective column is $400 a month, $4,800 a year, buying nothing. The genuinely new column is $400 a month, and that is the only part where the description matches something that did not exist three years ago.
Total defensible spend on the new scope: $400 a month out of $2,400. Your split will differ. The sorting takes twenty minutes and the columns are usually lopsided in the same direction.
Two honest notes on that arithmetic. Line items are rarely billed evenly, so ask for the hours behind each one rather than assuming. And the renamed column is not worthless work, it is work you may already be buying twice.
05Why does the renaming work so well?
Because it arrives with urgency attached and no way for the buyer to check.
SEO has a credibility problem earned over twenty years, so a new acronym is commercially useful. It lets a seller distance themselves from that history while selling the same services, and it lets them tell a business owner that everything they bought before is now obsolete.
It also exploits genuine uncertainty. The systems are new, they behave oddly, and nobody outside these companies can describe exactly how they choose sources. Uncertainty is the ideal condition for confident pricing.
Watch how quickly old manipulation tactics reappear under the new banner. Stuffing keywords into a business name was always against the rules, and you can report a competitor doing it today just as before. Buying fake mentions is the same move as buying fake reviews, and spotting a review you believe is fake uses the same instincts. New label, same behavior, same outcome.
06Does it matter if it is renamed, as long as the work happens?
Sometimes not. Often yes.
It does not matter if you get sound SEO, priced reasonably, described in whatever vocabulary the market currently prefers. The work is what you are buying.
It matters when the renaming causes double billing, meaning you already pay for content and technical work and now pay again under a second heading for the same activity.
It matters when it displaces effective work. An hour spent on a file that nothing fetches is an hour not spent fixing a Business Profile category.
And it matters when it sets expectations that cannot be met. AEO sold as a fast track to being named by assistants will disappoint, because the inputs are the slow ones: documentation, consistency, reviews and real mentions.
Here is where my own argument breaks down, and you should know it. Two of the three genuinely new items are real specialist work, and dismissing the whole category as renaming lets a lazy provider off the hook in the other direction. Running a measurement program at the sample sizes the research prescribes is not free. Neither is monthly log verification. A provider charging for those, and showing you the run counts and the log extracts, is charging for something that exists. The test is not the vocabulary. It is whether they can show you the artifact.
The good news for a business with limited budget is that the fundamentals cover most of it. Even a business with no website has real options in AI search, which tells you how much of this is about assets you already own.
07What to do this week
1. Print your current SEO or AEO proposal and mark every line as renamed SEO, genuinely new, or named as ineffective by Google. 2. Total the three columns in dollars, using the arithmetic above. The third column is the refund conversation. 3. Ask your provider which items would change if the AI features disappeared tomorrow. The honest answer is very few, and a good provider will say so. 4. For anything in the genuinely new column, ask for the artifact. Run counts per prompt, the prompt list itself, or a log extract with verified IPs. No artifact means no work. 5. Check that you are not paying two vendors for the same activity under two names. 6. Read Google's guidance yourself. Fifteen minutes, and it changes every conversation you have afterwards.
If a partner or client asks you what your position on all this is, what to tell someone who asks about AI search gives you a version you can say out loud without overpromising.
Be honest with yourself
When you do not need this
If you are not paying anyone for AI search work, you do not need to audit an invoice you do not have. Do the ordinary local work.
If your provider already tells you plainly that this is mostly SEO, they have passed the test this article is designed to run. Keep them.
And if you are a provider yourself, reading this as an attack on the vocabulary, it is not. Use whatever word gets the client to say yes to work that helps them. Just do not charge twice for it, and do not sell the four things Google names as ineffective.
Sources
- Google Search Central, "Optimizing your website for generative AI features on Google Search". Published May 2026, updated 10 July 2026. First party platform documentation. Source of the "still SEO" quote, the "AEO/GEO hacks" instruction, the third party tools warning, and the four named ineffective tactics.
- Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, "GEO: Generative Engine Optimization," arXiv:2311.09735. Accepted to KDD 2024. 10,000 query benchmark, tested against a simulated engine fed the cleaned text of five fixed Google results. Peer reviewed, and not validated for local intent.
- Schulte, Bleeker and Kaufmann, University of St. Gallen, "Don't Measure Once: Measuring Visibility in AI Search (GEO)," arXiv:2604.07585. 8 April 2026. Four engines, four verticals, a 45 day window plus same day repeated runs. Academic, primary.
- Ahrefs, "We Analyzed 137K Sites: 97% of llms.txt Files Never Get Read". June 2026, updated August 2026. 137,210 domains, log analysis by user agent and response code. Vendor research: Ahrefs sells SEO software, and states its own figures are ceilings.
Related reading
- What Google's own documentation says about optimizing for AI. The primary source every AEO proposal paraphrases, in one page you can read in fifteen minutes.
- llms.txt: what a 137,000 site study found. The full trace on the single most common ineffective line item.
- Measuring AI visibility without fooling yourself. What the genuinely new measurement work looks like when it is done properly.
- AI crawler user agents and what your server logs reveal. The other genuinely new item, and the artifact to ask a provider for.
Got a proposal with AEO on it?
Send me the line items from any AEO or AI search proposal at eric@seod.com and I will mark each one as renamed SEO, genuinely new, or named as ineffective by Google, with the reasoning next to each. No charge, no pitch, and if the proposal is good I will tell you to sign it.
I do this because the sorting takes me ten minutes and saves owners from paying twice for work they already have.
Otherwise, there is more on AI search and AEO here.