Generative engine optimization has moved from a research paper to a boardroom argument. The premise is simple: ChatGPT, Google AI Overviews, Perplexity and Copilot now answer many searches directly, so a brand has to get into the answer, not just onto page one.
What GEO actually is
GEO means shaping content, site access and off-site presence so generative AI systems retrieve, cite or name you when they compose an answer. That is a different prize from ranking in a list of links. The term dates to a November 2023 arXiv paper from IIT Delhi and Princeton researchers, and it picked up commercial speed after Google launched AI Overviews in May 2024 and OpenAI launched ChatGPT search in October 2024.
How answer engines pick sources
Answer engines run a retrieve-then-generate sequence. A page first has to be accessible and indexed. Then the model rewrites the query, pulls sources, and synthesizes an answer.
The competition is brutally concentrated. In 27 ChatGPT conversations examined by consultant Suganthan Mohanadasan in July 2026, the first search named a brand the user never typed in 21 cases. Brands named in those queries reached the answer 68.9% of the time; brands merely retrieved but not named appeared just 2.1% of the time. Only 3.1% of 3,554 retrieved pages were cited at all. Semrush’s data across 126 million US prompts put ChatGPT at 15.4 sources per response, against 3.3 for Gemini.
Why marketers should care
The click is the casualty. In the first randomized field experiment on AI Overviews, hiding the feature lifted outbound organic clicks from 0.37 to 0.62 per search. Publishers lose pageviews, but they are also the source material engines cite. Brands face a visibility squeeze too: Semrush tracked more than 1,200 brands and found only 36 held top-100 visibility on all four major platforms every month.
Still, terminology lags. In a July 2026 Fractl survey of 343 US marketing decision-makers, 19% had adopted the GEO label internally, while 81% still called the work SEO.
What to do about it
The tactics that show some signal fall into three buckets:
- Content: add statistics, quotations and attributed sources so pages give an engine something concrete to cite.
- Off-site: earn coverage, reviews and listings on the third-party pages engines actually retrieve.
- Technical: keep AI crawlers unblocked, render cleanly, and treat llms.txt or schema as optional rather than a magic switch.
Google’s own guidance, published May 15, 2026, insists generative AI search is still SEO. Brendon Kraham, a Google ads VP, has told CMOs “Good SEO is good GEO”. But Microsoft Advertising’s retail playbook separates answer engine optimization and GEO, and monitoring vendors such as Profound, Peec AI and Semrush are building a market around the distinction.
The honest catch
The evidence is thinner than the hype. The founding paper tested rewrites on a small pipeline. C-SEO Bench, later accepted at NeurIPS, found most methods largely ineffective and sometimes harmful to rankings. Self-promotional “best of” listicles have triggered what appears to be a Google spam action in January 2026, and a September 2026 Scrunch study found a cited self-promoting listicle only lifted its own recommendation rate from about 4% to 7%. A competitor was recommended instead in 24.3% of responses, and no brand at all in 63.9%.
The practical takeaway: treat GEO as a citation problem, not a new bag of tricks. If a page would not convince a careful editor, it will not convince an answer engine for long.
Source: PPC Land
