By Felipe Almeida · September 2, 2026
Answer engine optimization —AEO, or answer engine optimisation if you write in British English— is the work of getting an answer engine to name your brand when someone asks it a question and doesn't know yet that you exist.
That last part is what almost nobody measures, and it's the only part that matters.
When you ask ChatGPT "what does SearchBrand do?", ChatGPT answers well. That measures something else: that the model can read a website the user already knew about. The question that decides whether your AEO strategy works is the other one, the one that doesn't name you: "what platforms are there to monitor my brand's visibility in AI?".
We measure both separately, with 156 questions running daily against five models: ChatGPT, Gemini, Perplexity, Claude and AI Overviews. Here's the cut from September 2, 2026:
| Question type | Runs that day | SearchBrand appears | searchbrand.ai cited as a source |
|---|---|---|---|
| Brand named in the prompt | 140 | 137 (98%) | 115 |
| Brand not named in the prompt | 640 | 8 (1.3%) | 19 |
That gap —98% against 1.3%— is the one that matters, and it's exactly the one that disappears when a dashboard hands you a single visibility percentage. The weighted average of the two halves can read as 40% and it describes neither situation.
The diagnostic on our own domain returned the same reason on 22 of 30 opportunity cards: the engine's search worked, but it found no indexed content on searchbrand.ai. The questions were answered on the blog. The crawler couldn't see them.
Three causes, all of them implementation and none of them writing:
useEffect. renderToString doesn't run effects, so FAQPage, Article, BreadcrumbList and ItemList only showed up after hydration, when the bot was long gone.We fixed all three on August 28. Six days later, searchbrand.ai shows up as a cited source in 19 of the 640 daily runs of questions that don't name us, and as a brand mention in 8. That's still small for the size of the blog. What matters for anyone reading this is where the movement came from: not from writing one more article, but from three rendering fixes.
They get used interchangeably and they're worth separating, because they're measured differently:
In practice, AEO and GEO are the same discipline under two names with different origins. If someone sells you both as separate services, they're selling you the same thing twice.
What genuinely changes is the relationship with the click. Our own Search Console, three months through August 28, 2026: question-shaped queries ranked better than traditional keywords —median position 9.4 against 16.0— and converted ten times worse: 5 clicks on 49,528 impressions. The answer got resolved upstream.
That's the price of admission. Appearing in the answer doesn't bring you the visit; it brings you the recommendation. They're two different businesses and they need two different dashboards.
There's no official documentation, so this comes from looking at 2.1 million citations captured in our database. The patterns that repeat:
It retrieves before it answers. ChatGPT and Perplexity run a search, read a handful of pages and write from that. If your page isn't indexed or doesn't answer the question in plain text, it never makes the shortlist.
It prefers sources already shaped like an answer. The domains that show up most in our Spanish-language questions are agency and marketing-school blogs —iebschool.com in 142 runs, blog.hubspot.es in 176, cyberclick.es in 108— not the tools' own product pages.
It leans on directories. G2 shows up as a citable source in the AEO category. Our G2 profile exists, it's in the right category, and it has zero reviews. The models cited it four times anyway.
It deduplicates press releases. Four Chilean outlets published our story with nearly identical headlines in November 2025. The models weight that as a single mention.
Four numbers, in this order:
Appearance rate in non-brand questions. Runs where you get named, over total runs. That's the numerator that counts. If it's at 0, everything else is decoration.
Citation rate for your domain. How often the model uses your URL as a source. It moves before the mention does and it's the early signal that the technical work landed: in our case citations (19 of 640) run ahead of mentions (8 of 640).
Position inside the answer. Showing up fifth in a list of five isn't showing up. Right now in our non-brand questions, Profound appears in 185 runs with an average position of 4.1 and HubSpot's AEO tool in 152 with 3.3.
Sentiment and role. Being mentioned as "the cheap alternative" or as "the option for Latin America" changes the sale. It's text, not a number, and you have to read it.
What doesn't help: the aggregate visibility percentage, and Search Console impressions from prompt-shaped queries. Those second ones arrive in bursts of days and switch off just as fast. On August 30 we lost 850 impressions from that cluster in two days without moving a single position: Google simply stopped serving the query. If your dashboard mixes that with real demand, you'll celebrate and panic over nothing.
First, whatever unblocks indexing, which is the real bottleneck on almost every site built with a JavaScript framework:
curl your own URL and look for the answer text. If it isn't there, no crawler sees it.FAQPage, Article, BreadcrumbList. In React, with Helmet or equivalent, never inside an effect.And one format decision almost nobody makes: if the model's answer already resolves the question, your page needs to offer something the model can't replicate. A filterable table, a price with a verification date, a measurement of your own. Longer prose doesn't compete.
We don't recommend it if your problem is demand and not visibility. If nobody is asking questions where your category could show up, AEO work doesn't create that market. Same if your site isn't indexed in Google yet: the right order is indexing first, mentions second.
And there's a limit worth saying out loud: full model training depends on new versions, and that takes months. What moves in weeks is browsing —what the model reads at the moment it answers. Anyone promising to get you into the model in 30 days is talking about browsing and should say so. The same applies to Google AEO: AI Overviews reads the live web, it doesn't wait for a retrain.
What is AEO in search?
It is the work of getting an answer engine — ChatGPT, Gemini, Perplexity, Claude, AI Overviews — to name your brand when it answers a question. It is measured in mentions and citations, not in rankings.
Is AEO the same as GEO?
In practice, yes. GEO is the more common name in English and puts the emphasis on generative summaries. The discipline and the tactics are the same.
How long does an AEO strategy take to show results?
Technical indexing changes show up within days. On searchbrand.ai, six days after fixing three rendering bugs the domain was already cited in 19 of 640 daily runs. Brand mentions in questions that do not name you move slower, in weeks, and depend on content that actually answers the question existing.
Can you do AEO without tools?
You can prompt the models by hand and write down the results. It stops scaling around 20 questions, because each one has to run across several models and several times — the answer changes between runs. With 156 questions and 5 models that is 780 runs a day.
Does Google have its own AEO?
Not as a product. What exists is AI Overviews and AI Mode, which are answer surfaces inside Google. Search Console shows you impressions, but it does not tell you whether your brand was mentioned inside the answer.