How to Get Cited by ChatGPT, Gemini and Perplexity in 2026

The short version

A meaningful share of your buyers have stopped typing queries into a search box. They ask an assistant, get a synthesised answer with three or four citations, and click one of them — or none.

That makes citation share a channel. It is measurable, it responds to specific changes, and almost nobody is optimising for it deliberately yet.

What we tested

Across 40 client sites over 11 months, we changed one variable at a time and tracked citation frequency across a fixed prompt set. Four things moved the needle. Most of the received wisdom did not.

1. Entity clarity beats keyword targeting

Language models resolve entities, not strings. If your site never states plainly what you are, where you operate and what you sell, the model has to infer it — and inference is where you get dropped in favour of a competitor who was explicit.

The fix is unglamorous: a clear About page, consistent naming, Organization schema with a stable @id, and the same description of your business everywhere it appears.

2. Original data gets quoted; opinion gets ignored

This was the largest single effect we measured. Pages containing a number that exists nowhere else — a survey, an internal benchmark, a analysis of your own account data — were cited far more often than pages of well-written commentary.

Assistants are looking for something to attribute. Give them a fact only you have.

3. Answer the question, then elaborate

Content written to rank buries the answer under 400 words of preamble. Content that gets cited leads with the answer in the first two sentences, then earns the rest of the reader's attention.

This is also just better writing, which is the useful part: the same change improves the page for humans.

4. Freshness signals matter more than they do in classic search

Assistants weight recency heavily on anything that looks time-sensitive. A page dated 2023 competing with an equivalent page dated 2026 loses, almost regardless of quality.

What did not work

Keyword density. Word count. Publishing frequency. Stuffing FAQ schema onto pages with no real questions. Adding "2026" to titles without updating the content underneath — which, if anything, damaged us once the model noticed the mismatch.

Where to start

Pick your ten highest-intent pages. Make each one state plainly what it is about in the first paragraph. Add one piece of information that only you have. Get your Organization and Author markup right once, site-wide. Then measure monthly.

That is most of the work. The rest is patience.