Tideflow AI
Start free
Tideflow AI
Start freeLog in
← Blog

Blog · Tideflow AI

Do You Still Need Traditional SEO for AI Search?

Yes: traditional SEO is still the foundation for appearing in AI search, but it only gets you into the pool of sources an AI system chooses from.

Gabriel Baciu portrait
Gabriel Baciu
Published Updated
A translucent glass layer rests atop a solid stone foundation slab, symbolizing new methods built on established groundwork.

10 min read

Yes: traditional SEO is still the foundation for appearing in AI search, but it only gets you into the pool of sources an AI system chooses from.

Keep doing the fundamentals: crawlable pages, clean indexing, strong rankings, useful content, and real authority. Then add two things SEO never required. First, measure AI citations and mentions across repeated runs on each platform instead of checking rank once. Second, earn a presence in the third-party sources AI systems like to cite. You don't need to replace SEO with a separate discipline. You need to extend it.

Why rankings still decide who AI can cite

Most AI search products, including Google AI Overviews and AI Mode, ChatGPT search, Perplexity, and Gemini, work by running a web search and then summarizing what comes back. If your page doesn't turn up in that retrieval step, the model has nothing to cite.

The data reflects this. In a pre-registered study of 100,411 citation events across 2,000 queries and 14 industries, Google organic rank was the strongest per-page predictor of whether ChatGPT, Perplexity, Claude, or Google AI Mode cited a page. Pages in the top 3 were 7.8× more likely to be cited than pages ranked 11–30 (Anthony Lee, 2026). A companion study found that URLs in position 1 were cited by at least one AI platform about 54% of the time, compared with about 2% at position 100. Every platform showed the same steep drop (Anthony Lee, 2026).

A separate researcher reached the same conclusion independently. In a smaller study of ChatGPT and Gemini, rank was the dominant predictor. Position-1 pages were cited in 43% of the queries where they appeared, and position-7 pages in 5% (Fischman, 2026).

Two caveats apply. These are observational preprints rather than peer-reviewed papers, and they show correlation. Rank may be standing in for relevance and authority rather than directly causing citations. Either way, the practical conclusion holds: the pages that win in organic search are the pages AI systems cite most.

The "most AI citations go to pages that don't rank" claim is misleading

You may have seen the claim that most AI citations point to pages outside Google's top results, which would suggest rankings don't matter. The underlying number is real. In the large study above, 75.4% of citation events went to pages outside Google's top 30 (Anthony Lee, 2026).

The conclusion drawn from it is wrong. Consider a lottery. A huge crowd holding one ticket each will produce more total winners than a small group holding a hundred tickets each, yet every person in the small group has far better odds. Pages beyond the top 30 work the same way. There are vastly more of them, so they collect many citations in total, but each individual page rarely gets picked. In that study, most of these deep-tier citations were one-offs scattered across roughly 43,000 URLs and 16,000 domains.

For your own pages, the relevant number is the per-page odds. Those fall steeply as rank drops.

Where SEO stops working on its own

Ranking well gets you considered. It doesn't guarantee you get chosen. Even position 1 goes uncited roughly half the time or more, depending on the study. Three patterns explain the gap.

AI cites your domain, but not always the page you optimized. Alignment with Google was much higher at the domain level than at the URL level. The specific cited page often wasn't in Google's top 3 for the literal query (Anthony Lee, 2026). One likely reason is that AI systems rewrite the user's question before searching. This means depth across your whole site matters, not only whether one target page ranks for one keyword.

AI looks further down the results than people do. A peer-reviewed study found that, on average, 53% of the domains AI Overviews consulted weren't in Google's top 10, and 27% weren't in the top 100 (Kirsten et al., 2026).

Some engines barely search at all. In the same study, for identical queries, one GPT-4o tool configuration consulted less than one web page on average. Perplexity's Sonar consulted 14, AI Overviews and Gemini 9 each, and GPT search 4 (Kirsten et al., 2026). When an engine answers mostly from what the model already knows, your ranking matters less. What matters more is how widely and consistently your brand is described across the web. These configurations have probably changed since, but the spread shows that each engine rewards different things.

This is also why a citation and a mention are different signals. An engine can recommend your brand without linking to your site.

What AI search actually adds to SEO

The work that goes beyond SEO falls into three areas.

1. A different measurement problem

Rank tracking assumes the results page stays fairly stable. AI answers don't. When the same queries were run two months apart, only 18% of the web pages AI Overviews drew on appeared both times, compared with 45% for organic results. Even five minutes apart, answers changed in up to 27% of cases (Kirsten et al., 2026). Platforms also disagree with each other. Agreement across all four platforms in one study was close to random (Anthony Lee, 2026).

This doesn't make AI answers impossible to measure. Brand recommendations within a single platform were fairly consistent, with ChatGPT's run-to-run overlap around 0.62 on the Jaccard similarity scale. The fix is sampling: run the same prompts many times, on each platform, and track your share of answers over time. A single screenshot of one ChatGPT answer tells you almost nothing.

How you collect the data also matters. Reddit held 38.3% of Google top-3 positions in one sample but received zero AI citations through the platforms' APIs. The same platforms' web interfaces cited Reddit 8.9–15.6% of the time (Anthony Lee, 2026). If your monitoring only queries APIs, it may not reflect what real users see.

2. More weight on third-party sources

Platforms differ sharply in the sources they trust. Among lower-ranked citations, user-generated content made up 24% of Perplexity's citations and 0.6% of Claude's (Anthony Lee, 2026). Engines that retrieve little depend on what's already said about you elsewhere. Getting included in reviews, roundups, and community discussions that AI systems already cite is part of the job. For more on where that becomes legitimate strategy versus rebranded link building, see whether third-party listicles are good GEO or just AI-era link building.

3. Crawler access on its own terms

Retrieval works differently from platform to platform. In server-log testing, ChatGPT and Claude fetched pages live during conversations, while Perplexity and Gemini relied on pre-built indexes (Anthony Lee, 2026). About 1% of citations in the larger study went to URLs Google had dropped from its index entirely (Anthony Lee, 2026). Google's index isn't the only way in, so robots.txt rules and server reliability for AI crawlers deserve their own check.

Don't put AI-specific tricks ahead of the basics

The evidence for "GEO" content tweaks is thin. In the large pre-registered study, a composite of seven such features added only a small effect on top of rank, mostly driven by schema markup (Anthony Lee, 2026). The smaller study found that pages cited by AI and pages that weren't cited used generic schema at almost identical rates: 43.1% versus 44.8%. The one exception was Product or Review schema with concrete attribute fields filled in. Pages with that markup were cited 61.7% of the time versus 41.6%, and the advantage was strongest for lower-authority sites (Fischman, 2026).

The evidence on schema is unsettled. Detailed product and review markup is worth adding if it fits your pages. Treating schema as the key to AI visibility is not supported. For the broader argument about whether this counts as a new field, see Is AEO really a new discipline?

What to do this week

  1. Check your SEO foundation first. Pick the 10–20 questions buyers ask before choosing a product like yours. Find where you rank for each. If you're outside the top 10, fixing that is still your highest-leverage AI move.
  2. Run those same questions through each AI platform, several times. Ask each question at least five times on ChatGPT, Perplexity, Gemini, and Google AI Mode. Record whether you're mentioned, whether your site is cited, and who appears instead.
  3. Compare cited pages against the pages you rank. If AI cites a different page on your domain, read it. That page may answer the question more directly than your target page.
  4. List the third-party sources cited instead of you. Recurring roundups, review sites, and forum threads become your outreach list.
  5. Confirm AI crawlers can reach you. Check robots.txt and server logs for AI user agents such as OAI-SearchBot, ChatGPT-User, ClaudeBot, and PerplexityBot.

For a step-by-step version focused on one platform, see how to get mentioned by ChatGPT.

Keeping the signals separate

Step 2 gets tedious quickly. Five prompts across four platforms, repeated weekly, turns into hundreds of manual checks. The results are also easy to misread when different signals get blended together. A crawler hitting your site, an AI citing you, and a visitor converting are three separate events. Seeing one doesn't prove the others happened.

Tideflow AI is built around that distinction. It monitors the prompts your customers ask across ChatGPT, Gemini, Perplexity, Google AI Mode, and Google AI Overviews. It turns the gaps it finds into content plans, which can include owned articles or pitches for inclusion in third-party sources AI already cites. It also tracks AI crawler activity through bot detection. Tideflow treats a crawler visit as a discovery signal, not proof of indexing or inclusion in an answer, and a matching user agent as an observation rather than a verified identity. It doesn't guarantee citations or recommendations, and neither monitoring nor publishing can. Tideflow is currently in early access. If you're comparing tools, the Tideflow AI vs Peec AI comparison explains why a visibility score shouldn't be confused with traffic.

Frequently Asked Questions

Since ChatGPT reportedly uses Bing, should I optimize for Bing instead of Google?

Not at Google's expense. Citations on every platform studied, including those reportedly using Bing, matched Google's top results 4–7× more closely than Bing's (Anthony Lee, 2026). This is a correlation across roughly 100 queries and doesn't prove which backend any platform uses. It does suggest Google rank works as a good general proxy for the relevance and authority AI systems reward.

Will AI search reduce clicks to my site?

The evidence is mixed. A large causal study from before AI Overviews found that answer-style features such as featured snippets reduce click-through and keep people on the results page longer (Gleason et al., 2023). A lab study of Google's AI summaries found no significant change in engagement or traffic, though people paid more visual attention to the summaries (Rolon-merette, Guo et al.). Plan for some pressure on clicks, and measure mentions and conversions alongside traffic.

Sources

Explore more in Blog →