How to Rank in AI Overviews, ChatGPT, and Every AI Search Engine (2026 Data)
There's no single 'AI SEO' algorithm. Promptwatch's 26B+ citation dataset shows how ChatGPT, AI Overviews, AI Mode, Perplexity, Claude, and Copilot each decide who gets cited, and how to show up in all of them.
"Rank in AI search" is not one project. Promptwatch's public dataset, 26B+ analyzed citations, prompts, and responses pulled from the real interfaces of ChatGPT, Gemini, Perplexity, Claude, and Google AI Overviews, shows six engines with six different citation behaviors. ChatGPT leans on Reddit and increasingly on your own product pages. AI Overviews and AI Mode lean on Google's own properties. Perplexity is almost mechanically consistent. Copilot's citation count still swings by a factor of ten week to week. Optimizing for one and assuming the others follow is how teams end up "doing GEO" for months with nothing to show for it.
This is the map: what the data says each engine actually does, where the common ground is, and where you need an engine-specific play.
The citation math is different on every engine
Before formatting a single page, know how many seats exist. Promptwatch's average-sources-per-response data puts hard numbers on it:
| Engine | Avg. sources per response | Behavior |
|---|---|---|
| ChatGPT Search | ~5 | Smallest, most contested citation pool |
| Google AI Overviews | ~10 | Roughly double ChatGPT's slots, steady over time |
| Perplexity | ~10 | Remarkably consistent, barely moves day to day |
| Microsoft Copilot | 2 to 17, highly volatile | Still being re-architected by Microsoft; unreliable to plan against |
Fewer slots means fiercer competition per slot. ChatGPT citing roughly half the sources Google AI Overviews does is the single biggest reason a page that gets cited in one engine can be invisible in the other, it's not that your content failed, it's that ChatGPT only had five seats to give out and something more specific won them. Perplexity's steadiness makes it the best engine to A/B test content changes against, since a shift in your citation count there is much more likely to reflect your edit than engine noise.
The format shift: product pages are overtaking listicles
The clearest cross-engine trend in the data right now is a swing toward brand-owned commercial pages. In ChatGPT's July 2026 citation-type data, product pages hit 32.8% of citations, nearly double their ~18% share in March, while listicles sit at 9.7%. Google AI Overviews is on the same trajectory with a lag: listicles still average 18% for the month (down from ~26% in Q1), but product pages overtook them daily from July 28 onward, ending the month at 17.9% versus 16.2% for listicles.
Two engines, independently, moving the same direction is a trend, not noise:
- Your own product/pricing pages are now a citation asset, not just a conversion page. Structured specs, current pricing, and stock/availability data give models something concrete to lift.
- Listicles and "best of" roundups still matter, they're still the largest single format on AI Overviews and the second-largest on ChatGPT, but they're losing share every month, so don't build a strategy that depends entirely on winning a spot in someone else's roundup.
- Video is a rising, mostly-unclaimed format. Video citations on AI Overviews climbed from ~2.7% in January to nearly 6.3% by late July, and most brands still have nothing competing for those slots.
Reddit dominates the social layer, but not evenly
Promptwatch's social citation data confirms Reddit is the single biggest social source for AI search overall, but "biggest" hides how unevenly it's distributed:
| Engine | Reddit share | #2 social source |
|---|---|---|
| ChatGPT | 5.19% | LinkedIn, 0.23% |
| AI Overviews | 4.08% (YouTube) leads | Reddit close behind |
| Perplexity | 2.67% (YouTube) leads | Reddit close behind |
| Grok | 4.85% (YouTube) leads | Facebook 3.01%, Instagram 1.7% |
ChatGPT is a Reddit specialist, over 20x its next-best social platform. Google-adjacent engines and Perplexity are more evenly split between Reddit and YouTube. X barely registers anywhere (it never breaks 0.25%, even on Grok). If your GEO plan is "post more on social," match the platform to the engine your buyers actually use, or you're optimizing for citations that were never available to you.
Same platform, different engine, completely different rules
The clearest proof that there's no universal "AI SEO" is Promptwatch's LinkedIn citation breakdown, one platform, four engines, four different winning formats:
| Engine | What LinkedIn content it cites most |
|---|---|
| ChatGPT | Company pages (23.84%) and the LinkedIn homepage (22.55%), treats LinkedIn as a company directory |
| Google AI Mode | Pulse articles (44.85%), wants long-form, bylined expertise |
| Google AI Overviews | Pulse articles (42.25%), plus LinkedIn's own "top content" pages (8.14%) |
| Perplexity | Ordinary feed posts (41.88%) beat Pulse articles (32.46%), rewards fresh, specific takes |
A polished company page wins on ChatGPT and does close to nothing for AI Mode. A well-written Pulse article does the opposite. If you only maintain one LinkedIn asset, you're already choosing which engines you can show up in.
New crawlers are rewriting the access layer
Two things happened in the crawler logs in the last few months that change who can even compete:
- Claude is emerging as a citation source. Promptwatch's crawler data shows Claude's citation crawler going from ~30 visits a day in December 2025 to a peak of 1.73% share of all tracked AI crawler traffic by mid-April 2026, a 100x+ increase in four months. It's still small next to ChatGPT, but the growth curve means a robots.txt block that was harmless in December is quietly cutting you out of a fast-growing channel today.
- Meta may be building its own search index. Meta-WebIndexer's crawl share jumped from ~2% to 37.8% of all tracked AI crawler requests between mid-July and early August 2026, a 17x surge in under a month, lining up with reports that Meta wants to stop depending on Google. Whether your content was crawled during this build-out will determine whether it can ever be cited in Meta AI search.
Check your logs and robots.txt for Claude-User, ClaudeBot, meta-webindexer, meta-externalagent, and OAI-SearchBot before you do anything else. None of the content work below matters if the crawler can't reach the page.
One myth to retire: markdown does not help you get cited
Promptwatch tested this directly: across 1.67M citations from ChatGPT, Claude, Perplexity, and AI Overviews, HTML pages took 99.94% of citations and markdown (.md) files took 0.05%. AI search engines cite rendered web pages, not markdown mirrors. Markdown matters for AI coding agents fetching docs (Claude Code, Codex), which is a completely different, developer-facing retrieval channel. If your GEO plan includes publishing .md versions of blog posts "for the LLMs," that effort is going into a channel that carries essentially zero AI search citation weight, put it into your HTML instead.
What actually holds across every engine
Despite all the divergence, four things move citations everywhere in the dataset:
- Extractability. A clean, liftable claim in the first 40 to 80 words beats a well-written introduction, on every engine tested.
- The long tail has real odds. ChatGPT's June 2026 citation share shows the #1 domain (Reddit) holding under 4% of all citations, there's no winner-takes-all dynamic like classic Google rankings.
- Freshness doesn't mean recency. Both Reddit and YouTube citations skew old, 60%+ of cited Reddit posts and 72%+ of cited YouTube videos were over a year old at the time they were cited. Engagement (upvotes, likes) barely correlates with citation odds; topical precision does.
- Volatility is the norm, not the exception. Reddit's ChatGPT citation share dropped ~40% in a single month (6.11% to 3.71%, May to June 2026). A one-time audit tells you almost nothing about next month.
How to actually run this as a program
- Pick your target engines based on where your buyers actually search, not all six evenly. Read the ChatGPT deep dive, Google AI Overviews and AI Mode deep dive, and Perplexity, Claude, and Copilot deep dive for the engine-specific plays.
- Fix crawler access first, allow OAI-SearchBot, Claude-User/ClaudeBot, and Meta's indexing bots unless you have a specific reason not to.
- Format for the shift: tighten product/pricing pages, keep a handful of tight comparison pages, and don't over-invest in listicle placements as your only lever.
- Track it daily with evidence. None of this is visible in Google Search Console or any classic analytics tool. Promptwatch checks ChatGPT, Gemini, Claude, Perplexity, Copilot, and Google AI Overviews every day, logging citation and crawler data behind every tracked answer, with a weekly action digest email flagging notable swings, which is how you catch a swing like Reddit's 40% drop the week it happens instead of in a monthly report.
FAQ
Is there one "AI SEO" checklist that works for every engine?
No. The data above, different citation counts, different social platforms, even different winning formats on the same LinkedIn page type, shows engine-specific behavior is the rule, not the exception. Common fundamentals (extractability, crawler access, freshness) matter everywhere; format and channel choices don't transfer.
Which engine should I prioritize first?
Whichever one your buyers actually use to research your category. If you don't know, Promptwatch tracking your prompt list across its full engine set for two weeks will show you where you already have traction versus where you're invisible.
Do AI Overviews and AI Mode need separate strategies?
Largely yes, see the dedicated guide for how differently they weight Google's own properties versus the open web.
How often does this data change?
Frequently enough that a single snapshot is dangerous. Reddit's ChatGPT share moved 40% in a month; ChatGPT's site: operator usage jumped 46x in a single day on August 8, 2026. Promptwatch's public data reports update regularly, and your own tracked prompts should be checked daily, not quarterly.
What to do this week
- Check your robots.txt and server logs for OAI-SearchBot, ClaudeBot/Claude-User, and Meta's indexing crawlers. Unblock anything you didn't mean to block.
- Audit your top 10 commercial pages for a liftable answer block, current pricing, and structured product data.
- Pick the two or three engines that matter most for your buyers and read the matching deep dive.
- Put 25 to 40 real prompts into a daily tracker. Promptwatch Essential covers a broad engine set and citation analytics at $95/mo. Professional adds 25M AI crawler logs at $245/mo. Either way, the next shift in this data shows up in your dashboard instead of three months later in a client meeting.