Get cited by ChatGPT & Perplexity: 2026 GEO playbook

Getting cited by ChatGPT or Perplexity is decided passage by passage, not page by page: the model lifts a self-contained chunk that already carries a clear claim, a number and an attribution, so the job is to rewrite existing sections into that shape rather than to publish more pages. The Princeton/Georgia Tech GEO study (arXiv 2311.09735) reported that adding quotations lifted visibility in generative-engine answers by 42.8% and adding statistics by 33.2%. The qualification: those are aggregate lifts measured across many queries, not a guarantee for any single one.

How to get cited by ChatGPT and Perplexity: a tactical GEO playbook

Getting cited comes down to rewriting a passage so it stands alone: one claim, one number, one attribution, formatted so the model can lift it without editing. The Princeton/Georgia Tech GEO study tested nine optimisation methods applied to existing source pages and found quotation addition lifted visibility by 42.8%, statistics by 33.2% and source citations by 27.8%, while keyword stuffing cut it by 8.7%. That ranking is the rewrite priority order below.

The rewrite sequence

Work section by section on pages that already rank, not on new pages. For each H2 or H3:

1. Isolate the claim. Cut the paragraph to one sentence that answers the heading directly — no throat-clearing, no "in this section we'll discuss."
2. Attach a number. If the paragraph has no figure, find one (internal data, a named source, a dated stat) or drop the paragraph — vague claims don't get lifted.
3. Name the source inline. A citation inside the sentence ("...according to [X]...") outperforms a footnote; models preserve attribution when they lift text.
4. Add one direct quote where credible — from a named expert, a customer, or a documented result. Quotation was the single highest-lift tactic in the Princeton/Georgia Tech data.
5. Cut the connective tissue. Delete transition sentences between the fact-bearing lines; each sentence should survive being extracted alone.

Before / after example

Before (page as it ranks today)	After (rewritten for citation)
"Many businesses are now looking at AI search as an important part of their strategy, and it's worth considering how your content performs there."	"Adding direct quotations to a page lifted its visibility in AI answers by 42.8% in the Princeton/Georgia Tech GEO study (arXiv 2311.09735), the largest gain of the nine methods tested."
"Our support team is very responsive and helps customers quickly."	"Our support team resolves [your measured figure]% of tickets within one business day (source: your helpdesk report, [month/year])." — a template, not a claim: fill it with a figure you can produce on request, never an estimate.

The one-passage test decides citability, not overall page length or keyword density. Before publishing a rewrite, read the target paragraph in isolation and ask whether it would still make sense, and still carry a number and a source, if pasted into a chat window with no surrounding context.

Run this rewrite pass across an existing topic cluster before writing anything new. Those pages already carry the site's authority signals, and a cluster gives a model several adjacent passages on the same subject to choose from.

What actually makes a passage worth quoting?

Our working rule: a liftable passage answers one question completely in 40–80 words, states a checkable number or named fact, and attributes it — so ChatGPT or Perplexity can copy the sentence without pulling in the rest of the page for context. That self-containment is the whole game. The model isn't reading your page; it's scoring a chunk against a query and deciding whether that chunk stands on its own.

Retrieval-augmented generation operates on passages, not pages. In broad terms, an answer engine splits a document into chunks, ranks each chunk against the query, then quotes or paraphrases whichever one needs the least surrounding context to make sense. This is why the Princeton/Georgia Tech study tested discrete content edits rather than whole-page rewrites — the unit of optimisation is the passage.

That study found two edits dominate: adding direct quotations lifted visibility by 42.8% and adding cited statistics by 33.2%. Source citations added a further 27.8%. Only one tactic backfired — keyword stuffing cut visibility by 8.7%, the single negative result across all nine methods tested.

The practical implication: write the named fact and its source in the same sentence as the claim, not in a footnote three paragraphs later. A passage that says "X reduces Y by 12%, according to [source]" is liftable; a passage that says "many studies show significant improvements" is not, no matter how well the surrounding page is written.

What is GEO, and how is it different from ranking?

GEO (Generative Engine Optimisation) means writing and structuring content so an AI answer engine can lift a passage and attribute it — the win condition is a citation inside a synthesised answer, not a ranked position. Traditional SEO optimises for a spot in a list; GEO optimises for being one of the few passages an LLM decides is quotable enough to paraphrase or cite by name. We set out what AI search rewards in more depth on our Generative Engine Optimisation page, and how it differs from AEO and classic SEO in AEO vs GEO vs SEO.

The mechanics force this difference. Google hands a user ten blue links to sort through themselves; ChatGPT and Perplexity instead hand back one synthesised answer built from a small set of quoted or paraphrased sources — a handful of domains per answer rather than a page of results. So the real target isn't "rank higher"; it's "be one of the few passages judged good enough to quote."

That changes what "coverage" means in practice. A page ranking #4 on Google still gets clicked; a page that's well written but not extractable gets read by the crawler and never surfaced in the answer at all. The practical unit of success shifts from page rank to passage-level extractability. A single self-contained paragraph with a checkable fact, not the page as a whole, is what gets lifted and attributed, which is why a GEO workflow has to be built around passages, not URLs.

How do you rewrite an existing page to be citation-ready? (step-by-step)

Rewrite a page section by section rather than starting over: take a URL that already ranks or draws traffic, and rework each H2 against the four checks below before touching anything else. Most citation-ready assets on a mature site started as an ordinary page edited into shape, not a fresh publish. That matters because you already have the topical authority signal and backlinks; what's missing is retrieval structure.

1. Isolate one question per section. If an H2 currently answers two questions — say, "what it costs and how it works" — split it into two H2s. A passage that drifts from its heading loses the standalone quality that retrieval systems need to lift it cleanly out of context.

2. Open with the direct answer. Rewrite the first sentence of each section to state the claim in full: the actual number, name or mechanism, not "there are several factors to consider." The Princeton/Georgia Tech GEO study measured quotation addition as the single strongest optimisation lever tested, lifting visibility by 42.8%, which is consistent with models rewarding passages that front-load a quotable, self-contained claim.

3. Attach a named, checkable fact. Every rewritten section needs a statistic, a study name, a spec, or a dated data point the model can quote without needing to verify it elsewhere. The same study found statistics added 33.2% and source citations 27.8% to visibility — unsupported claims tend to get paraphrased into vagueness or dropped entirely, not cited.

4. Add the attribution inline. Name the source next to the figure it supports, in the sentence itself, rather than in a footnote or a closing "sources" list. That is what lets the attribution travel with the fact when a model paraphrases the passage.

Work through a page's H2s in this order, ship the edits individually, and hold new pages to the same bar. That is the structural check every drip-published article at IT Master goes through (how it works), so the rewrite pass and the new-content pipeline enforce the same standard.

What structural cues tell ChatGPT and Perplexity a passage is safe to quote?

A retrieval system judges a passage safe to quote when the heading states the user's exact question, the paragraph beneath it answers that question directly within one or two sentences, and the source of the claim is named inline — not by schema markup or a target word count. These are pattern-matching decisions made at the passage level, so structure inside the passage matters more than anything wrapping it.

Heading-passage match is the strongest of the three signals. When a heading reads as a real question — "How do I get cited by Perplexity?" rather than "Citation Strategy" — the retriever can score the heading and the sentence beneath it as a matched pair, instead of scanning several paragraphs to guess where the answer starts. A mismatch, where the heading is generic and the answer arrives three sentences later, forces the model to infer relevance rather than confirm it, which lowers the odds that passage gets pulled into a synthesised answer.

Inline attribution does similar work for trust scoring. A sentence that names its source ("according to the ICO's guidance" or "per the vendor's own documentation") reads as verifiable in a way an unsourced assertion doesn't, and citing sources is one of the levers the Princeton/Georgia Tech GEO study found measurably shifted visibility, alongside adding direct quotations and statistics.

Short paragraphs finish the job: two to three sentences per idea means the retriever can lift the whole unit without truncating mid-thought, which is exactly the failure mode that makes a quoted passage look unfinished or wrong.