To optimise content for ChatGPT, Perplexity, Claude and Google AI Overviews you need to accept that they retrieve differently, favour different sources, and will cite the same page at wildly different rates. Overlap between platforms is poor, so a single “AI-optimised” template gets you cited by one engine and ignored by the others. This is what each one does, which structural patterns pay off across all four, the specific edits I make to a draft, and the order to work in when you can’t do everything.
The most useful fact in this whole area is that an analysis of 680 million citations found only 11% of domains get cited by both ChatGPT and Perplexity, and a 2026 study of 34,234 responses found a 46-times gap in brand citation rates between platforms. Those two numbers kill the idea of generic AI optimisation. To optimise content for ChatGPT is a different job from optimising it for Perplexity, and both differ again from Claude and AI Overviews.
The strategy behind it, meaning why substance comes before structure, sits in my AEO/LLMO Content Optimisation Framework, and the general citability mechanics are in the tactical guide to AI-citable content. This piece is about what changes per platform.
How do ChatGPT, Perplexity, Claude and AI Overviews cite differently?
Four different retrieval methods, which produce four different source preferences.
| Platform | How it retrieves | What it favours | What that means for a draft |
|---|---|---|---|
| ChatGPT | Selective search activation, with a large share of answers coming from training data instead of live retrieval. | Consensus and reference sources, established publishers, and visible credentials. | Be the consensus, not the contrarian, on the definitional parts, and put credentials and named authority on the page. |
| Perplexity | Live web search on effectively every query, with retrieval and rerank. | Community sources, with Reddit at around 46.7% of top citations, and recent material. | Freshness and dating matter most here, and your presence in actual communities does work your website can’t. |
| Claude | Cites the documents in front of it, and is around 30% more likely to cite bullet-pointed pages. | Depth and clean structure. | Structure the depth instead of cutting it, since this is the one engine that rewards long, well-organised pages. |
| Google AI Overviews | Google’s existing ranking infrastructure, with roughly 54% overlap with organic rankings. | What already ranks, though frequently from outside the top three results. | Your existing SEO work is the input; this is the least novel of the four. |
Two consequences worth sitting with:
- Freshness is a platform-specific lever, not a universal one. Content updated recently is cited at materially higher rates on Perplexity, where recency is part of the ranking logic, while ChatGPT’s cycle-based indexing weights stability more. So a quarterly refresh with a visible update date is cheap, and it targets one engine specifically.
- ChatGPT is the hardest to win and the biggest traffic source. Slate HQ’s study of 300,000+ citations across six B2B SaaS brands found Claude gave brands the highest owned citation share at 9.1% and ChatGPT was consistently the worst, while Conductor’s benchmark put 87.4% of AI referral traffic through ChatGPT. Low citation share on the platform sending most of the traffic is an awkward combination, and it’s why a single share-of-voice number misleads you.
Which structural patterns get cited across all four?
These go into every draft, whether I’m trying to optimise content for ChatGPT, Perplexity or neither in particular.
- Explicit definitions at first use. Every key term defined in a sentence that survives being lifted out of the page entirely. If the definition only makes sense inside its paragraph, no engine is extracting it.
- Question-format headings with self-contained answers. The H2 is the question a person asks, and the 40 to 75 words underneath answer it completely before any context arrives. Around 44% of LLM citations come from the first 30% of a page, so burying the answer is expensive.
- Comparison structured as a table, by which I mean an actual comparison with values in the cells and not ticks, which hands a model discrete facts with the relationships already resolved. This is the pattern that pays off most consistently, and it’s why the comparison page is such an under-built asset.
Underneath all three sits the substance requirement. The Princeton GEO study (Aggarwal et al., ACM KDD 2024, tested across 10,000 queries) found attributed statistics lifting visibility around 41% and external citations up to 115% for lower-ranked content, while adding words did nothing. Structure delivers the substance underneath it and can’t stand in for it, which is the argument in why the AEO playbook is usually run backwards.
What do the edits really look like?
Specific changes, not principles.
A buried definition, front-loaded
Before: “Over the past few years, as buying committees have grown and procurement has become more involved, many teams have found that the traditional approach to lead scoring no longer serves them well.”
After: “Lead scoring assigns a numerical value to a prospect based on their behaviour and profile, to decide who sales contacts first. It works poorly for committee purchases, and here’s why.”
The second version defines the term, states a position, and survives extraction on its own. The first says nothing a model could lift.
An unsourced claim, attributed
Before: “Most B2B content never gets used.”
After: “Forrester’s finding is that 60 to 70% of content produced by B2B marketing organisations goes unused, with some companies reporting above 80% after that figure was published.”
Same claim. One version is an assertion, the other is a citable fact with a named source, which is the single biggest lever in the research.
Prose comparison, converted to a table
Any paragraph containing the word “whereas” plus two product names is a table that hasn’t been built yet. Pull the attributes out, make them rows, put a value in every cell for every option, and date the whole thing.
What should you prioritise when you can’t optimise for everything?
In this order, which reflects payoff per hour and not the sequence most guides use. Nothing below matters if you skip the first item, including everything above about how to optimise content for ChatGPT specifically.
- Substance that isn’t available elsewhere. Original data, a named source nobody else has quoted, a position somebody could disagree with. Every platform-specific tactic below is worthless applied to a page a model can reconstruct from ten others.
- Front-loaded answers and explicit definitions. Cheap, and they work on all four engines.
- Comparison tables where a comparison exists. Highest-return structural change, and it doubles as decision content.
- Whatever your buyers actually use. Check your own referral data and ask in sales calls before assuming ChatGPT. A research-heavy B2B audience skews toward Perplexity in ways a consumer audience doesn’t.
- Freshness and update dates. Quarterly refresh with a visible date, aimed squarely at Perplexity.
- Community presence. Not a website change at all, which is why it gets skipped, and yet it’s the direct route into Perplexity’s preferred source pool.
- Schema and technical markup. It matters, it’s hygiene, and it goes last, because it’s the layer everybody does first and it explains the least variance.
The one I’d fight for hardest is number four. Optimising for the platform your buyers don’t use is a well-executed waste, and the way you find out is running a prompt inventory across engines and logging who gets cited.
The editorial checklist
Run this on every draft before it publishes. It applies whether the target is to optimise content for ChatGPT or any of the other three, and no item takes longer than a couple of minutes.
- [ ] Does the page contain at least one claim, number or named source a model can’t get from ten other pages?
- [ ] Is every key term defined at first use, in a sentence that works out of context?
- [ ] Do the first 40 to 75 words after the H1 answer the title question completely?
- [ ] Is every H2 a question somebody actually asks, phrased the way they’d ask it?
- [ ] Does each section stand alone if it’s the only part extracted?
- [ ] Is every external figure sitting immediately beside its source link?
- [ ] Is any genuine comparison built as a table with values, not ticks?
- [ ] Is there an FAQ block of three or four self-contained pairs?
- [ ] Does the page carry a visible last-updated date, and is it true?
- [ ] Is every claim traced to documentation, data or a named stakeholder in the source log?
- [ ] Do related pages link to each other instead of repeating each other?
The last two carry more weight than the formatting items above them. One fabricated statistic costs the authority signal on the whole page once models cross-reference claims across sources, and a cross-linked set of pages reads as depth in a way that fifteen orphaned posts never will.
Honest about the limits
Every number in this piece comes from a vendor study, and the vendors sell AI visibility tooling.
They also disagree with each other. One index puts Wikipedia at 47.9% of ChatGPT’s top citations, another at 7.8%, which is a gap too big to be a rounding difference and probably reflects different query sets and different definitions of a citation. The 11% overlap figure and the 46-times variance are consistent across several sources and I’d defend those. The precise source-share percentages I wouldn’t.
The Princeton numbers are from 2024 and the engines have changed since. What’s held up across everything I’ve applied this to is the ordering: substance, then structure, then measurement. The platform-specific tactics sit on top of that and they’ll date faster than the rest of this page will, which is also why llms.txt is worth less than the discourse suggests and why the investment call between SEO, AEO and LLMO is a sequencing decision and not a technology bet.
Frequently asked questions
Partially. The shared layer works on both: front-loaded answers, explicit definitions, question-format headings, comparison tables, and attributed data. The platform-specific layers pull in different directions, since Perplexity weights recency and community sources heavily while ChatGPT weights consensus, established reference sources and visible credentials, and only around 11% of domains are cited by both. The practical approach is to build the shared layer into every piece, then pick one platform-specific lever based on where your buyers actually research.
Less than the other engines do, since AI Overviews run on Google’s existing ranking infrastructure and overlap with organic rankings by roughly half. Cited pages frequently come from outside the top three results, so ranking on page one matters more than ranking first, and the extractability work (self-contained sections, definitions, direct answers in the opening) is what converts a ranking page into a cited one. Existing SEO investment is the input, not a separate project.
It depends on the audience, and checking beats assuming. ChatGPT carries the overwhelming majority of AI referral traffic, at around 87% in Conductor’s benchmark, while giving brands the lowest citation share of the major engines, so it’s simultaneously the biggest channel and the hardest to win. Perplexity cites brands far more readily and skews toward research-heavy buyers, and Claude gives the highest owned-brand citation share of the platforms studied. Run your own prompt set across all four and read your referral data before committing effort.
Quarterly, with a visible and accurate update date, since recency is a ranking factor on the engines that search live for every query and content refreshed within the last three months is cited at meaningfully higher rates than stale pages. The update has to be substantive, meaning new data, corrected claims, or a revised position, because changing a date on unchanged content is the kind of signal that costs credibility when a model cross-references it against other sources.
