Skip to content
AI for Business

AEO Strategy for Websites: Stop Ranking, Start Getting Cited

AEO strategy for websites goes beyond SEO structure. Real agent readiness, crawler rules, and citation traps most guides skip - plus what actually moves the needle.

6 min readBeginner

Your best page just vanished into an AI answer

Last spring I watched a client lose a $12k deal without a single bounce on their site. The prospect asked Claude for the best mid-market inventory tool for food distributors that handles lot tracking. Claude named two competitors. Never us. Our page sat at #3 for the exact phrase. Clean schema. Polished FAQ. Zero citation.

That loss ended the “add question headings and FAQ schema” playbook for me. A lot of AEO strategy for websites is still featured-snippet SEO in a newer hat. Ranking puts you on a list. Getting cited means the model trusts a passage enough to lift it into the answer. Different job.

If the next buyer starts in an assistant instead of Google, the only score that matters is whether the agent names you. Here’s the rebuild after that miss.

What AEO is actually for

AEO means your copy can be extracted, trusted, and named inside replies from ChatGPT, Claude, Perplexity, Google AI Overviews, and agents built on them – not just ranked for a click. Cloudflare puts it bluntly on their AEO overview: the shift is from “rank so a human clicks” to “be found, read, and recommended by the system guiding the human.”

As of 2026, Cloudflare Radar shows fewer than half of HTML page requests come from people. Agents already show up as traffic. They don’t scroll for fun. They grab clean passages, check provenance, synthesize. You can own the blue link forever and still stay invisible in the answer if the useful bit is buried – or if the crawler never got a machine-readable copy.

Practical setup: can the agent even read you?

Almost every beginner guide jumps to headings. I didn’t. First question after the lost deal: did any retrieval bot see the page?

Cloudflare’s free isitagentready.com scanner (and the Agent Readiness tab in the dashboard) turned that from vibes into a punch list. They shipped the diagnostics in April 2026 and expanded AEO Visibility – citation rate, prominence, mention rate, share of voice versus competitors – in August 2026, probing models like Claude and GPT with category prompts. Details live on their Agent Readiness write-up.

Scan the domain. Checks cluster by effort:

  • Quick wins: robots.txt with explicit AI crawler rules, XML sitemap, clean Markdown alternative for agents
  • Technical groundwork: Content Signals, Link headers, API catalog
  • Advanced: MCP/A2A agent cards, OAuth discovery, Web Bot Auth

Biggest payoff for most sites? robots.txt. Training bots and live retrieval bots are not the same animal. OpenAI’s own bot docs split them.

# Allow live search/retrieval (citations)
User-agent: OAI-SearchBot
Allow: /
User-agent: ChatGPT-User
Allow: /
User-agent: Claude-SearchBot
Allow: /
User-agent: Claude-User
Allow: /
User-agent: PerplexityBot
Allow: /

# Decide on training separately
User-agent: GPTBot
Allow: / # or Disallow if you refuse training data use
User-agent: ClaudeBot
Allow: /
User-agent: Google-Extended
Allow: / # Gemini/AI training signal; does not replace classic Googlebot

Blocking GPTBot does not auto-block ChatGPT answers. OAI-SearchBot and ChatGPT-User handle the live fetch path. I watched a team “protect” the site by disallowing every AI user-agent, then stare at empty Perplexity results for weeks. Selective allows fixed it the same day logs showed the right bots returning.

Pro tip: After robots.txt changes, re-scan Agent Readiness and read raw server logs. Mis-labeled scrapers still show up; verify IPs when you can.

Then shove the answer into the early page. Cite.solutions tracking (as of their 2026 AEO 101 notes) puts roughly 44% of AI citations in the first 30% of the text. Open with a 40-60 word block that stands alone on the core question. No throat-clearing intro.

Extraction-ready copy that survives synthesis

Reachable isn’t enough. Each major section has to survive being snipped out of context.

Question-form H2s help matching – sure. The sentence under the heading still has to answer the whole thing. Short paragraphs. Parallel lists only when the comparison is real. Tables when numbers beat prose. Named sources and figures so the model has ground to stand on.

FAQPage JSON-LD? Only on real visible Q&A. CapstonAI-style 2026 cohorts floated citation lifts around 1.9× on ChatGPT up to ~3.1× on AI Overviews; Relixir-type snapshots claimed similar gaps (e.g. 41% vs 15%). An Ahrefs-style controlled run on nearly 1,900 pages landed near zero – or a small drop. The split usually comes down to markup that mirrors the HTML plus evidence-dense answers, not thin filler. Still run Google’s Rich Results Test. FAQ rich results themselves are mostly gone for everyday sites; engines can still use the structure for understanding.

Off-page punches above its weight on checklists. Reddit threads, review roundups, third-party write-ups feed pools models already trust. One useful reply in the right subreddit moved more citations for us than three fresh blog posts.

Freshness stacks on all of it. An 18-month-old URL can keep classic rankings and go quiet inside answer engines; recent signals weigh heavy.

Honest limits

Measurement is messy on purpose. Commercial AEO dashboards often hit APIs or lab sessions that don’t match what a person sees in the ChatGPT or Claude UI. Same prompt five times, different citations. Geo, personalization, model version – more noise. A lone “17% share” printout is a weather report, not a KPI. Cloudflare’s AEO Visibility at least multi-probes and reuses category baselines; still a snapshot.

JS-heavy pages keep biting people. Answer only after client hydration? Plenty of crawlers never see it. Server-render the critical blocks, or ship the Markdown negotiation path Agent Readiness looks for.

You can’t force a cite either. Authority, consistent brand facts across the web, and passage quality still pick winners. AEO sits on strong SEO. It doesn’t replace it.

FAQ

Do I need both SEO and AEO?

Yes. Drop either and you lose.

How do I know if AI is recommending competitors instead of me?

Grab 10-20 real buyer questions from sales calls and tickets. Paste them into the major assistants. Log who gets named. Last quarter a client did this on Monday morning and found two rivals owning “lot tracking for distributors” while their #3 page never appeared – same gap that cost the $12k deal. Cloudflare’s AEO tab (category probes, share of voice; expanded August 2026) automates some of that. Re-run weekly. Single runs lie.

Is FAQ schema still worth it after Google gutted FAQ rich results?

For classic SERP snippets, value fell off a cliff. For answer engines, it can lower extraction friction when the markup mirrors visible, useful pairs – and when the answers aren’t fluff. It is one signal. Not a switch. Dense evidence beats the script tag every time; the cohort-vs-controlled split above is the proof.

Open isitagentready.com, paste the homepage, clear every red quick-win before you rewrite another H2. That afternoon of agent readiness work is still the highest-impact AEO move most sites haven’t made.

Related Tutorials