Ranking in ChatGPT, Google AI Overviews and Perplexity takes three separate playbooks, because each surface retrieves sources differently. This guide covers how to rank in AI Overviews, ChatGPT and Perplexity one surface at a time: what to publish, where to earn mentions, and how to verify the work moved your recommend rate.
The tactics below rest on first-party data. On 13 July 2026 we at Honeyb (this is our product) ran 20 buyer prompts, three identical times each, across four engines via API: GPT-5 mini, Gemini 2.5 Flash, Claude Haiku 4.5 and Perplexity Sonar. That produced 240 answers and 2,507 extracted citations to compare.
Why one playbook fails across three surfaces
Each engine builds answers from a different citation pool. In our July test, ChatGPT cited 900 sources across 445 distinct domains, and its three most-cited domains earned just 7.6% of all citations. That is a long tail. Perplexity drew 14% of its 498 citations from Reddit and another 8% from YouTube. Gemini, the closest measurable sibling to AI Overviews, wrapped 74% of its URLs in Google's own grounding redirects, a clear signal that it retrieves through Google's index. Only one domain, Forbes, appeared in all four engines' top citation lists.
| Surface | Retrieval pattern | Key numbers (Honeyb, 13 July 2026) | Priority tactic |
|---|---|---|---|
| ChatGPT | Long-tail web search | 445 domains cited; top 3 domains take 7.6% | Publish specific, structured pages |
| AI Overviews | Google's index | Gemini routed 74% of URLs via Google grounding | Get indexed, answer questions directly |
| Perplexity | Web plus community sources | Reddit 14% and YouTube 8% of citations | Earn thread and video mentions |
Sources per answer
Average sources cited per answer, by engine
Citation width shapes strategy. ChatGPT averaged 15.0 sources per answer in our test, against 10.7 for Gemini, 8.8 for Claude and 8.3 for Perplexity. A wider net means more slots to occupy, which is why ChatGPT rewards breadth of coverage while Perplexity rewards presence in a small set of community sources.
How to rank in ChatGPT
ChatGPT is the most democratic of the three surfaces. With 445 domains sharing 900 citations and no dominant publisher, a focused page on a niche buyer query has a realistic shot at retrieval. Publish pages that settle one comparison or one decision each: pricing breakdowns, head-to-head tables, honest capability lists. Structure beats prose. Open with a direct answer, then give the model tables it can lift.
Target the long tail deliberately. ChatGPT's brand-set overlap between identical runs was just 42% in our data, the lowest of the four engines, which means it samples widely rather than repeating a fixed list. Each distinct page covering a distinct buyer question is another ticket in that sample.
Earning mentions matters as much as publishing. Ahrefs found AI visibility correlates most strongly with third-party mentions and video, not on-page work. Semrush adds that 62% of AI citations never name the brand, so aim to be named in the body text of other people's content: reviews, comparison posts, tool directories. Our guide on how to get cited by AI covers the outreach mechanics.
How to rank in AI Overviews
Want to see this in action?
See how every major AI model talks about your brand. Free to start.
AI Overviews is the most conventional surface of the three because it draws on Google's index. If Google cannot crawl and rank a page, AI Overviews cannot summarise it. Standard technical hygiene still applies: indexable pages, fast rendering, headings that mirror the question being asked. Our Gemini data points the same way, with 74% of cited URLs passing through Google's grounding redirects rather than the open web.
The tactical difference from classic SEO is answer shape. Put the direct answer in the first sentence under a question-form heading, keep it tight, and support it with a table or list. We cover eligibility, formatting and markup in detail in how to appear in AI Overviews.
How to rank in Perplexity
Perplexity is the community surface. In our test it took 71 of its 498 citations from Reddit (14%) and 40 from YouTube (8%), and Semrush's larger study puts Reddit at 40.1% of all AI citations, the single most-cited source. You cannot buy your way into those threads. Answer real questions in your category's subreddits under a disclosed affiliation, and treat detailed, useful replies as content assets rather than promotion.
Video is the second lever. A plain walkthrough or comparison video becomes a citable YouTube source, and the Ahrefs correlation data suggests video mentions carry more weight than another post on your own domain. Perplexity also exposes its sources on every answer, so you can audit exactly which threads and channels it trusts for your queries, then target those.
The measure, fix, remeasure loop
None of this work counts until the answers change, and single spot checks mislead. In our test the top recommendation changed between identical runs 44% of the time on Gemini, 43% on Perplexity, 35% on ChatGPT and 28% on Claude. SparkToro reports the same instability, with the same AI query changing roughly 70% of the time. The honest success metric is recommend rate: the share of repeated runs in which your brand appears, measured before and after each change.
| Engine | Top pick changes between identical runs | Brand-set overlap between runs |
|---|---|---|
| Gemini | 44% | 54% |
| Perplexity | 43% | 61% |
| ChatGPT | 35% | 42% |
| Claude | 28% | 67% |
Run the loop in fixed cycles. Fix one surface at a time, hold your prompt set constant, and compare recommend rate against the baseline rather than any single answer. Honeyb, our platform, automates the daily runs from $29 per month for 10 prompts on one model (pricing checked 13 July 2026), and our guide to measuring AI share of voice explains the maths. Start with a free AI visibility check to get your baseline before you change anything.
