AI Search & AEO: Frequently Asked Questions

AI search isn't a future thing — it's already where a growing share of your customers ask their questions. They type into ChatGPT, ask Perplexity, or get an AI answer above Google's links, and a machine reads sources and writes the answer for them. I build an engine that does this for a living (Otto, the system behind RunOctopus), so these are the questions I get asked most, answered the way I actually understand them from shipping against real answer engines. If you want the full mechanics, start with the AEO playbook.

What is AEO (answer engine optimization)?

AEO is structuring your content and your site so AI answer engines retrieve it, ground their answers in it, and name you as a source. Three verbs, in order: retrieve, ground, cite. The goal isn't ranking a blue link a human clicks — it's being one of the small handful of sources the model quotes when it writes the answer. If any one of those three verbs fails, you're invisible in that answer.

Is SEO still worth doing in 2026?

Yes, absolutely. Search is being rewritten and expanded into AI answers, not deleted, and AEO sits directly on top of the foundation SEO has always built: crawlable pages, topical authority, and clear writing. A page no crawler can read and no one links to or trusts loses in both worlds at once. The honest framing is that SEO and AEO are complementary disciplines that share a spine — AEO just adds an extraction-and-markup layer on top of good SEO.

How do ChatGPT, Claude, and Perplexity decide what to cite?

Most of them run a retrieval step first. The system rewrites your question into one or more search queries, hits an index, and pulls back a set of candidate pages. Then it splits those pages into passages, scores each passage against the query, and feeds the best ones into the model, which writes an answer grounded in them with citations attached. You get cited when your page is retrievable, your passage directly and clearly answers the query, and your claim is specific enough to quote without the model having to hedge. Want to go deeper on this? I broke it down in get cited by AI search.

Does schema markup (JSON-LD) actually help?

It helps, but it's not a cheat code. Clean JSON-LD — FAQPage, Article, Organization, Product — makes your claims and entities unambiguous to machines and powers the rich results that feed AI overviews. It won't rescue thin or inaccurate content; no amount of markup makes a bad page worth quoting. But on a genuinely useful page, structured data removes friction between your answer and the engine trying to extract it, so it's worth doing right.

What is llms.txt and do I need one?

llms.txt is a proposed plain-text file you place at your site root that points AI systems to your most important, cleanest content — think of it as a curated map for machines. It's an emerging convention, not a confirmed ranking input: no major answer engine has promised to honor it yet. It's cheap to add and does no harm, so I ship one on sites I build. Just don't expect it to do the heavy lifting that genuinely good pages, clear structure, and authority do.

Should I block or allow AI crawlers?

If you want to show up in AI answers, you generally have to let the crawlers in — block them and you've opted out of the retrieval engines entirely. The nuance is that there are two kinds: crawlers that retrieve-and-cite live (these send you visibility and sometimes clicks) and crawlers that scrape purely for model training. Most businesses should allow the answer-engine crawlers without a second thought and only get selective about the training-only ones if they have a specific reason. Opting out of citation to protect against training is usually a bad trade for a business that wants to be found.

How do I tell if AI search is already citing me?

Test it directly — don't guess. Ask the real engines (ChatGPT with search, Claude with web access, Perplexity, Google's AI mode) the actual questions your customers ask, and see whether your domain shows up in the cited sources. Do it across a fixed list of prompts so you're measuring the same thing each time, log which prompts cite you, and re-run it monthly. That prompt-coverage rate — what fraction of your target questions name you — is your real AEO scoreboard, and it's far more honest than any vanity metric.

How long does AEO take to work?

Faster than classic SEO in some ways, slower in others. Live-retrieval engines like Perplexity can pick up a strong new page within days of crawling it, because they're searching a fresh index every time you ask. Models that lean on training-data memory move on a slower cycle tied to their training runs. In practice you start seeing citation pickup in weeks on the retrieval engines, with authority compounding over months as more of the web describes you consistently.

Does AI search kill my website traffic?

It reshapes your traffic more than it erases it. Some purely informational clicks now get answered inside the AI response and never reach your site — that's real. But being the cited source still drives qualified visits and puts your brand in front of the customer at the exact moment they're deciding. The pages that lose are thin ones that only ever existed to capture a click; the pages that win are the ones worth quoting. The move is to be the source, not to mourn the click.

What content formats get cited most?

Answer-first content, where a clear and self-contained claim sits right under a heading that matches the real question. FAQ pages, direct how-to steps, head-to-head comparisons, and crisp definition pages all extract cleanly because each chunk stands on its own. The format itself matters less than the structure underneath it: lead with the answer, make every section quotable in isolation, and be specific enough that a model can lift your sentence verbatim instead of paraphrasing a cleaner source.

Do backlinks still matter for AEO?

Yes, indirectly. Links and brand mentions remain part of how engines judge authority and trust, and that trust feeds whether your passage gets pulled and cited over a competitor's. But for AEO specifically, consistent and accurate descriptions of your business across the web — your entity authority — matter as much as raw link counts. If five sites describe what you do the same clear way, models absorb that; if every site describes you differently, the model stays fuzzy on you.

Can I just generate a thousand AI pages and win?

No, and it usually backfires. Answer engines are built to extract a trustworthy, specific claim, and a flood of generic machine-written pages gives them nothing quotable and can erode the trust signals you've earned. Volume without substance is the fastest way to look like the source a model skips. Depth wins: fewer pages that genuinely answer real questions beat a content farm every time. If you do build with AI, hold it to a real quality bar — I cover how in building with LLMs.

What's the single biggest AEO mistake people make?

Burying the answer. Most of us write to build up to a point — answer engines reward the exact opposite. If the answer to a specific question lives in paragraph nineteen, you'll lose to a page that answers it cleanly in the first two sentences under a heading that matches the question. The mechanical fact that changes everything: you're optimizing passages, not pages. Treat the first two sentences of every section as the thing the model will lift, and write accordingly.