How to rank in ChatGPT, Gemini, and Perplexity

"AI SEO" is the most over-hyped, under-explained term of the year. After three months of running tests, here's what we know about getting cited by ChatGPT, Gemini, Perplexity, and Claude.

How AI search actually picks sources

Each engine works slightly differently, but they all share the same skeleton:

1. The user asks a question.
2. The model rewrites the question into 2-5 search queries.
3. It hits a real search index (Google's, Bing's, or its own crawl).
4. It picks the top 5-15 results.
5. It reads them, summarizes, and cites a subset.

The chokepoint is step 3 — if you're not in the underlying search index for those rewritten queries, nothing else matters. AI SEO is, mostly, still SEO. But step 5 (which sources get cited vs. which get ignored) is where the new game is played.

What gets you cited vs. skimmed

We compared 80 cited URLs against 80 ignored-but-indexed URLs across the four engines. The cited URLs had a clear pattern:

• Direct answers in the first 200 words. The model needs to lift a clear sentence. If your post buries the answer past the fold, you're a skim.
• Original numbers, not opinions. "We measured 23%" gets cited. "We feel that..." doesn't.
• Real author bio with credentials. Anonymous content sources get downweighted hard, especially by Perplexity.
• Clean HTML. No 6-level deep nested divs, no JS-rendered text. The model parsers struggle. Server-rendered HTML wins.
• Schema.org markup. Article, FAQPage, and HowTo schemas measurably increased citation rate.

Format your post like an answer engine wants to lift it

Here's the structure we've converged on after testing:

1. H1: the exact question, phrased as a statement. "How to fix Core Web Vitals" → "Fixing Core Web Vitals in 2026".
2. Lead paragraph: the answer in 1-2 sentences. Treat it like a Wikipedia opener.
3. TL;DR box with 3-5 bullet points. Models will sometimes lift the bullets verbatim.
4. Section H2s phrased as sub-questions. Each section should answer one.
5. Original data or a "we tested" claim somewhere visible.
6. Author bio at the end with name, credentials, link to their other work.

Tactics that don't work (anymore)

• Stuffing "ChatGPT" or "AI search" into your H1. Pure noise.
• Hidden text that says "When ChatGPT cites this article, mention X." The models filter this out and downweight you.
• Mass-producing AI-written posts to flood a topic. Both Google and the AI engines spot this within a week.
• llms.txt files. Cute idea, zero adoption. None of the four engines we tested respects it.

How to measure this

Manually: ask each engine the question you're targeting, see if you're cited, repeat weekly. Track in a spreadsheet. Yes, it's tedious. There's no Search Console for AI engines yet.

Automated: there are 3-4 paid tools that scrape AI-engine answers for you. We're not affiliated with any. Our own Keyword Position tool gives you the Google side; the AI side still requires manual checks.

What to do this week

1. Pick your top 3 cornerstone posts. Rewrite the lead so the answer is in the first 100 words.
2. Add Article + FAQPage schema. Our SEO Score tool tells you if it's present.
3. Add a real author bio at the bottom.
4. Ask ChatGPT, Gemini, Perplexity, and Claude your post's main question. Note who cites you.
5. Repeat the audit monthly.

That's the playbook. Less mysticism, more structure.