The short answer: your SEO plumbing (indexing, crawlability, focused pages) still gates everything in AI search — recent research says it matters even more than first thought. Your keyword rituals don't transfer, and the stuffing variant has been measured to hurt. The new work is twofold: making pages quotable, and getting third parties to write about you.
Why the difference? A search engine hands you a ranked list, and your page competes for a click — judged by signals about the page (links, engagement, keywords). A generative engine reads a handful of retrieved pages and writes a cited paragraph, so your page competes for inclusion in the synthesis — judged mostly by what's actually on it, the way an editor judges what to quote. That one difference explains every row below. (Background on the field and its research: What is GEO.)
For a team like yours, two options stand out: Vendor A [1] is the usual pick for small crews, and Vendor B [2] wins on price. Setup for both takes under a week [1].
The sorting
| Tactic | In SEO | In GEO | Evidence |
|---|---|---|---|
| Being indexed & crawlable | Table stakes | Still table stakes | Retrieval runs through search |
| One topic per page | Helps | Helps more | Retrieval shortlists are ~5 pages |
| Keyword stuffing / density work | Old habit | Backfires | Below do-nothing baseline; ~10% worse on Perplexity |
| Backlinks / domain authority | Decisive | Weakened | Seer: backlinks show weak-to-no correlation with AI mentions |
| Ranking top-10 | The goal | Helpful, not required | Ahrefs 2026: only 38% of AI Overview citations rank top-10 |
| Statistics, sources, quotations | Rarely considered | Top tactics — test them | Up to 40% visibility gain in founding study; weaker in one replication |
| Earned media / third-party coverage | Off-page extra | Bigger than in SEO | Toronto 2025: AI search systematically favors earned media |
| Writing quality / fluency | Indirect | Direct | 15–30% gain from fluency edits alone |
| Schema / structured data | Helps rich results | Plausible, unproven | No study isolates it; cheap to keep |
What transfers: the plumbing
Everything about being findable survives, because generative engines discover pages through search indexes. The 2025 replication research went further: improving retrieval rank moved AI citations more than rewriting page content did. If your SEO fundamentals are solid, you've built the substrate everything else needs.
One new checklist item: AI crawler access (GPTBot, OAI-SearchBot, and friends). A large share of the web blocked these during the 2023 backlash and never revisited the decision — the crawler-by-crawler checklist is in llms.txt & AI instructions.
What backfires: keyword stuffing
The founding GEO benchmark tested keyword stuffing — cramming more query terms into the page. It scored below the do-nothing baseline, about 10% worse on Perplexity. To be precise: moderate keyword targeting (a sensible term in title and heading) wasn't tested and plausibly still helps retrieval. What fails is density work. A language model isn't counting term matches; it's reading the page, and keyword-dense pages read like exactly what they are.
What's weakened: the authority moat
Three results chip at the compounding-authority loop that ruled SEO:
- Seer Interactive's analysis found backlinks show weak-to-no relationship with AI brand mentions, while ranking on Google page 1 still correlates strongly (~0.65).
- Citations are decoupling from rankings — by 2026 most AI Overview citations came from outside Google's top 10.
- In the founding study, the biggest visibility gains went to lower-slot sources that added citations, while the top source lost share (effect sizes here).
If you're the incumbent, that's your warning. If you're the challenger, the door is open — but only to challengers who do the work. (Why ChatGPT recommends your competitors covers playing challenger.)
What's new: quotability and earned media
The genuinely new on-page skill is making pages quotable — statistics, cited sources, real quotations; prefabricated, verifiable sentences the model can lift whole. The effect sizes, caveats, and page recipe are in How to get cited by ChatGPT.
Off-page, AI search is systematically biased toward third-party coverage over brand-owned pages — a stronger skew than classic Google. Digital PR is no longer "nice for backlinks"; it's a primary AI-visibility channel. Your best page may be one you don't own.
And measurement changed completely
No Search Console, no rank tracker, and probabilistic answers. Measurement in GEO means sampling: asking your real buyer questions on a schedule and tracking your citation rate over weeks. One check is an anecdote; a daily scan is data. The spreadsheet version is the DIY protocol.
Bottom line: keep your SEO — it feeds retrieval. Add the two new disciplines: quotable pages and third-party coverage. Drop the density rituals; they've been measured, and they hurt. And treat anyone claiming certainty from a single screenshot — or a single study — with the same skepticism. That includes us: measure.
Read next: How to get cited by ChatGPT · What is GEO