The three acronyms describe overlapping work. SEO is the practice of earning rankings and clicks in a search results page. AEO (answer engine optimization) is the practice of being the answer an AI system gives when someone asks a question. GEO (generative engine optimization) is the practice of being included, cited, or recommended when a generative system composes a response. In practice AEO and GEO describe nearly the same job from slightly different angles: AEO focuses on the answer itself, GEO focuses on the engine that generates it. Both sit inside SEO rather than replacing it.
The short version
- SEO earns a position in a ranked list. The unit of success is a ranking and a click.
- AEO earns the answer. The unit of success is being the content the system states, quotes, or links when a question is asked.
- GEO earns inclusion in a generated response across many possible prompts. The unit of success is coverage: how many of the prompts your buyers actually type return you.
If you already do technical SEO, content strategy, and digital PR well, most of that work transfers. What changes is where you aim it and how you measure it.
What transfers from classic SEO
Crawlability and indexability
If a crawler can't fetch your page, nothing downstream happens. This is the least glamorous part of the comparison and the one that breaks most often.
AI systems use documented crawlers, and the rules differ by vendor:
| Crawler | What it does | Effect of blocking |
|---|---|---|
| OAI-SearchBot (OpenAI) | Surfaces websites in ChatGPT search | Sites opted out are not shown in ChatGPT search answers |
| GPTBot (OpenAI) | Collects content to improve foundation models | Blocking does not affect ChatGPT search |
| ChatGPT-User (OpenAI) | User-initiated actions in ChatGPT and custom GPTs | Because the user initiated it, robots.txt may not apply; not used to decide search inclusion |
| ClaudeBot (Anthropic) | Collects content that may be used for training | Blocking excludes future content from training |
| Claude-User (Anthropic) | Fetches pages when a user asks Claude a question | Blocking prevents retrieval for user queries |
| Claude-SearchBot (Anthropic) | Indexes content to improve search results | Blocking prevents indexing for search |
| PerplexityBot | Surfaces and links sites in Perplexity search results | Not used to crawl content for foundation models |
| Perplexity-User | Fetches pages for user questions | Generally ignores robots.txt because the user requested it |
| Google-Extended | Controls training of future Gemini models and grounding in Gemini apps and the Vertex AI Gemini API | Does not affect inclusion in Google Search and is not a ranking signal |
The practical takeaway: a blanket "block all AI bots" rule is a decision, not a default. Blocking GPTBot does not remove you from ChatGPT search. Blocking OAI-SearchBot does. Those are different choices with different consequences, and the vendor documentation is explicit about it.
Two operational notes worth knowing. OpenAI says robots.txt changes can take about 24 hours to take effect; Perplexity says the same. Anthropic says its bots honor robots.txt, support Crawl-delay, and publishes its IP ranges.
AI Overviews are a different case entirely. They are part of Google Search, so they are governed by normal Googlebot crawling and indexing. Google-Extended has no bearing on them.
Authority and being cited
Generative systems assemble answers from sources they can find and, where the product shows citations, from sources they choose to name. The signals that made a page citable in traditional search — original data, clear attribution, a named author, a real organization behind it — still matter. What's new is that a source can be selected for a short answer even when it doesn't rank first, and a page that ranks first can be skipped if the system prefers a cleaner explanation elsewhere.
Clear answers
Writing a direct, self-contained answer to a specific question was good SEO practice before AI answers existed. It's now closer to the core of the job, because a generated answer often pulls one or two sentences rather than sending a reader through your page.
What's genuinely new
You're optimizing for a slot, not a position
A ranked list has ten or more slots. A generated answer usually names a handful of sources, sometimes one. The competitive set is smaller and the selection logic is less visible. You cannot inspect a ranking algorithm for a generative answer the way you can inspect a SERP.
Source selection is partly a product decision
Which sources appear depends on what each product indexes, what its crawlers are allowed to fetch, and how it presents citations. Some of that is documented, as in the crawler table above. Some of it is not publicly documented, and anyone claiming to know the exact weighting is guessing. Treat confident explanations of "the AI ranking algorithm" with suspicion.
Prompt coverage replaces keyword coverage
Keyword research gave you a list of phrases with volumes. Buyers now type full questions, often with context ("best CRM for a 12-person agency already using HubSpot"). One page rarely covers that space. You end up mapping a set of prompts — the questions you'd want to be part of the answer for — and checking which ones return you. That's a different unit of work, and it's the part of GEO that isn't just renamed SEO.
Measurement is noisier
Rank trackers give you a stable number. Generative answers vary by phrasing, by session, and by product. Any measurement approach has to accept that and sample repeatedly rather than treat one result as the truth.
The comparison table
| Work | SEO | AEO | GEO | Transfers? |
|---|---|---|---|---|
| Technical crawlability | Core | Core | Core | Yes — same work, new crawler names |
| robots.txt policy | Standard | Affects retrieval and search indexing | Affects search indexing | Yes, but decisions differ by bot |
| Keyword research | Core | Partial | Partial | Becomes prompt research |
| Direct answers on page | Helpful | Core | Core | Yes, with more emphasis |
| Structured data | Rich results | Helps machines parse | Helps machines parse | Yes |
| Backlinks and mentions | Ranking signal | Indirect | Indirect | Yes, as credibility evidence |
| Original data and research | Differentiator | Differentiator | Differentiator | Yes |
| Earning a short-answer slot | Not applicable | Core | Core | New |
| Source selection by product | Not applicable | Core | Core | New, partly undocumented |
| Prompt coverage | Not applicable | Core | Core | New |
| Presence in generated recommendations | Not applicable | Partial | Core | New |
| Ranking position tracking | Core | Weak proxy | Weak proxy | Doesn't transfer cleanly |
Where GEO gets oversold
Three claims worth pushing back on.
"GEO replaces SEO." It doesn't. If your site can't be crawled, or your pages are thin, or nobody links to you, changing your copy to "answer-first" won't fix it. The crawl and authority layers are shared.
"There's a trick to getting cited." There are documented crawler rules and there are writing practices that make a page easier to quote. Beyond that, the selection logic inside each product is not fully public. No tool or consultant can guarantee a mention or a citation, and we don't either — anyone selling that is selling something they can't verify.
"You can measure it exactly." You can measure whether a set of prompts returns you, and you can track that over time. You cannot reduce it to a single stable rank, because the underlying outputs aren't stable. For a fuller treatment of what visibility means here, see What Is AI Search Visibility? A Plain-English Definition for Marketers.
A practical way to split the work
Keep your SEO program running. Then add three things on top of it:
- Audit your crawler rules. Decide per bot, not in bulk. Check whether OAI-SearchBot, Claude-SearchBot, and PerplexityBot can fetch the pages you want cited. Remember that changes can take about a day to apply.
- Build a prompt list. Write down 20 to 50 questions your buyers would actually ask a chatbot, in their words. Not keywords — questions.
- Check coverage, not rank. For each prompt, note whether you appear, whether you're cited, and who is named instead. Repeat monthly and look for patterns in the sources that show up.
That third step is where a tool like PromptRanking fits: it tracks whether ChatGPT, Gemini, and Perplexity recommend a brand, which is the coverage question rather than the ranking question. It reports what those systems return; it does not guarantee that any brand will be recommended. You can run the same check by hand for a small prompt list if you'd rather start there.
Next step
Open your robots.txt file and check it against the crawler table above. For each bot, write one line saying whether you want it fetching your site and why. If the answer is "we never decided," that's your first fix — and it costs nothing but an afternoon.