SEO people noticed the shift first. Their dashboards started softening last summer — not crashing, just thinning. Branded queries held up. Long-tail informational traffic started bleeding. The cause was not a Google update. The cause was that a meaningful slice of their audience had stopped typing into a search box and started typing into Perplexity, ChatGPT, or Gemini.
We have been calling this Generative Engine Optimization, mostly because nobody has come up with a better name. The acronym is GEO. It is not a replacement for SEO. It is a parallel discipline, with overlapping rules and a different goal: getting cited inside an AI-generated answer, not just ranked on a results page.
What is actually different about GEO
Three things have changed enough to matter. First, the unit of competition is the answer, not the page. An LLM stitches an answer from many sources and may surface only a small set of citations. You can rank #1 on Google for a term and not be cited at all by a Perplexity answer for the same intent. Second, the input the model retrieves from is messy and time-sensitive. Some engines lean on their own search index, some on Bing, some on a curated mix. The retrieval surface is fragmented in a way Google never was. Third, recency is weighted more aggressively. A two-year-old definitive guide can lose to a six-month-old blog post that uses cleaner formatting.
What is working in practice
1. Be the primary source, not the summary
Engines cite the page that originated the claim, not the page that re-explained it. We have stopped advising clients to write 'ultimate guides' that consolidate other people's research. We push them toward original data, original frameworks, and named methodologies they own. A page titled 'The 2026 SmartRecruiter Migration Benchmark, Mudish Technologies' will outperform a page titled 'The Complete Guide to ATS Migration' every time, even with worse on-page SEO.
2. Tighten the citation surface
Pages that get cited tend to share three traits: an unambiguous claim near the top, a number or definition the model can lift verbatim, and a clear publish date. We add a short stat or definitional sentence in the first 120 words of every commercial page. Schema helps but is not the lever — formatting is.
3. Treat each engine as a different surface
Perplexity is closest to traditional search and rewards link equity. ChatGPT browse leans on its own retrieval, which currently overweights well-structured Q&A and tables. Gemini surfaces YouTube heavily for product comparisons. Pretending these are interchangeable will leave you optimizing for none of them.
What does not work
- arrow_rightStuffing keywords. The retrieval layer is semantic. Synonym soup makes a page worse, not better.
- arrow_rightHidden 'AI instructions' in the page source. Engines either ignore them or downrank you for trying.
- arrow_rightRe-publishing the same article on Medium, LinkedIn, and Substack. Duplicate-content suppression has become aggressive in retrieval.
- arrow_rightLong, undifferentiated thought leadership. If the model cannot extract a sentence to quote, you will not be cited.
A 30-day GEO audit
- arrow_rightWeek 1 — Identify the 30 buyer-intent queries you most want to be cited for. Run each through Perplexity, ChatGPT, and Gemini. Log the cited sources.
- arrow_rightWeek 2 — For every page you would like to see cited, audit the first 120 words. Add a definitional sentence and a date. Surface the publication date in the URL slug or page metadata.
- arrow_rightWeek 3 — Replace any 'ultimate guide' page that does not contain original data with a narrower piece that does. One number is more powerful than ten paragraphs.
- arrow_rightWeek 4 — Stand up the citation tracker as a recurring report. Decide which engines you actually care about, and stop optimizing for the rest.
GEO will not replace SEO. The best clients we work with are running both, with separate dashboards and separate KPIs. The teams getting blindsided are the ones that thought their old SEO discipline would carry them. It will not. The next decade of organic discovery will be split between two surfaces, and the work to win each is genuinely different.
