AEO vs GEO vs SEO: What Actually Changes, and What Doesn't

Two engines, two answers, four months apart

If you have sat in a meeting where someone asked whether the company needs a GEO strategy on top of its SEO strategy, you already know the conversation goes nowhere. Half the room thinks it is a rebrand invented to sell retainers. The other half has seen ChatGPT recommend a competitor and wants someone to fix it by Friday.

Both instincts are defensible, which is why the argument never resolves. What is new is that the two largest players have now put their positions in writing, and the positions do not match.

That disagreement is more useful than either answer on its own. It tells you the question is badly formed. "Is GEO different from SEO?" has no single answer because it depends entirely on which engine you mean, and increasingly, which layer of the work you mean.

What each acronym actually means

SEO is the oldest and the broadest: earning visibility in a set of results a search engine returns. The unit of success is a position in a list.

AEO, answer engine optimization, is not new either, whatever the current marketing suggests. Its lineage runs directly through featured snippets, the "position zero" chase, and voice search. The unit of success shifts from being a result to being the answer. Anyone who spent 2017 restructuring pages so a definition sat in a clean 40-word paragraph under an H2 was doing AEO before the acronym existed.

GEO, generative engine optimization, has a precise origin that is worth knowing because it is routinely misquoted. The term comes from a 2023 paper by Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, later accepted to KDD 2024. The authors framed the problem honestly: generative engines are black boxes, and "content creators have little to no control over when and how their content is displayed."

The paper's headline finding, that GEO methods "can boost visibility by up to 40%," is the single most abused statistic in this field. Read the abstract carefully. Forty percent is an upper bound, not an average. It was measured on GEO-bench, a purpose-built benchmark, not on live Google. And the authors' own conclusion undercuts any universal playbook: "the efficacy of these strategies varies across domains, underscoring the need for domain-specific optimization methods." Anyone quoting the 40% figure without those three qualifications is selling something.

Google's position: it is all still SEO

Google published its generative AI optimization guide in May 2026 and has kept it current. It is the most direct statement any engine has made on this question, and it addresses AEO and GEO by name:

From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.

The mechanical reason follows in the same document. Google's AI features run on two techniques layered on the existing index. Retrieval-augmented generation, which Google also calls grounding, "relies on our core Search ranking systems to retrieve relevant, up-to-date web pages from our Search index." Query fan-out generates concurrent related queries to gather more results. Google's worked example: a user asking how to fix a lawn full of weeds triggers fan-out queries like "best herbicides for lawns" and "remove weeds without chemicals."

If retrieval runs on the Search index, then whatever gets you into the Search index is your AI strategy. Google states the eligibility rule explicitly: to appear in generative AI features, a page must be indexed and eligible to be shown in Google Search with a snippet, and the site must not have opted out via the Search generative AI features control in Search Console. We covered that control, including the inheritance trap that can silently exclude a subdomain, when Google published its AI search playbook.

Google also points site owners evaluating GEO vendors toward its guidance on third-party SEO tools and advice, which warns that third-party tools have no access to Google's internal ranking data and cannot guarantee performance. That is not subtle.

The mythbusting section every GEO pitch should have to survive

Google's guide includes a section titled "what you don't need to do." It is worth reading against any GEO proposal on your desk.

The field data on llms.txt is unflattering too. Common Crawl analyzed 584,107 llms.txt files in its July 2026 crawl and found roughly two thirds were generated by a plugin, only about half followed the structure the specification actually defines, and a few contained prompt injections. A standard that mostly gets installed rather than authored, and that half of adopters implement incorrectly, is not yet a standard.

One caution worth flagging: Google is describing Google. Microsoft's guidance in the same period is softer on structure, advising that clear headings, tables, and FAQ sections "help surface key information and make content easier for AI systems to reference accurately." These are not contradictory so much as differently weighted. Structured, well-organized content was already good practice for readers. It just does not need a machine-only version. The wider point, that Google's advice describes one reader among several, is the whole argument of why Google's AI advice does not travel.

Where the three genuinely diverge

Now the useful part. Three real divergences exist, and none of them are content tactics.

Crawler permissions are per-engine and not interchangeable

This is the most concrete difference, and it is the one most often skipped. OpenAI documents four user agents with genuinely different jobs. OAI-SearchBot governs whether you surface in ChatGPT's search features, and OpenAI is blunt about the consequence: "Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though can still appear as navigational links." GPTBot governs training. ChatGPT-User handles user-initiated fetches, and OpenAI notes that "because these actions are initiated by a user, robots.txt rules may not apply," and that it "is not used to determine whether content may appear in Search." OAI-AdsBot only visits pages submitted as ads. OpenAI also notes it can take roughly 24 hours for a robots.txt change to propagate to its search systems.

Anthropic separates its crawlers the same way: ClaudeBot for training data, Claude-User for content fetched at a user's direction. Anthropic also warns that IP blocking is not a reliable opt-out because it prevents robots.txt from being read, and that it does not publish IP ranges.

The practical upshot: "block AI crawlers" is not a coherent instruction. Blocking a training crawler is a licensing and IP decision. Blocking a search crawler is a visibility decision that removes you from an answer surface. Plenty of sites have made the second choice by accident while intending the first.

Measurement is per-engine and non-comparable

Google's generative AI performance report in Search Console gives impressions in AI features, the URLs that appeared, countries, devices, and date granularity. It does not give clicks and it does not give queries.

Bing's AI Performance gives almost the inverse emphasis: total citations, average cited pages per day, page-level citation counts, and grounding queries, described as "the key phrases the AI used when retrieving content that was referenced in AI-generated answers." Microsoft notes the grounding query data is a sample.

You cannot add these together. One counts impressions without queries, the other counts citations with sampled queries. Any dashboard presenting a single blended AI visibility score is manufacturing a number that does not exist. Track them as separate surfaces with separate baselines, the same way you would never blend paid and organic into one number and call it traffic.

The economics changed even where the tactics did not

This is the divergence that actually justifies giving the work its own line in a plan.

Pew Research Center found that users who landed on a Google results page with an AI summary clicked a result 8% of the time, versus 15% without one, based on browsing data from 900 US adults in March 2025. Sessions ended entirely on 26% of pages with an AI summary, compared with 16% without. We unpacked what that does to the old keywords-to-clicks staircase in the straight line to number one is gone.

Cloudflare's 2026 report puts numbers on the supply side. More than 50% of internet traffic is now non-human. As of June 2026, 52% of crawler requests are for AI training, up from 22% in spring 2025. Mixed-use crawlers, which blend search, agent use, and training, account for over 36% of activity, which makes the clean training-versus-discovery distinction above harder to act on than it looks. Cloudflare also reports that some of the most heavily crawled categories have seen human traffic decline as much as 40% in under a year, and that Google still accounts for roughly 88% of referral traffic. Its earlier work introduced the crawl-to-refer ratio, which measures how much a platform crawls relative to how much traffic it sends back.

Nothing in Google's mythbusting list is wrong. It is just answering a narrower question than the one a CMO is asking. "Do I need different content?" is mostly no. "Does the same content now produce fewer visits per unit of visibility?" is measurably yes.

What this means for how you run the work

Treat it as one discipline with three reporting surfaces and one new operational layer.

The content and technical work is shared. Google's advice to create non-commodity content with a first-hand point of view is the same advice that makes a page worth citing in ChatGPT. Google's example of the distinction is unusually concrete: "7 Tips for First-Time Homebuyers" is commodity content anyone could have written, while "Why We Waived the Inspection and Saved Money: A Look Inside the Sewer Line" is not. A model summarizing five interchangeable pages has no reason to name any of them. Note also that mass-producing pages for every query variant runs into Google's scaled content abuse policy, so the volume approach is both ineffective and risky.

What is genuinely additive is operational, not editorial: an audited robots.txt where every AI user agent is an explicit, intentional decision; separate baselines in Search Console and Bing Webmaster Tools; IndexNow for freshness on surfaces that reward it; and, for ecommerce, feed submission, since AI shopping surfaces increasingly gate on structured product data rather than page content.

One further layer is arriving. Google's guide now includes a section on agentic experiences, noting that browser agents may access your site by "analyzing visual renderings (like screenshots), inspecting the DOM structure, and interpreting the accessibility tree," and pointing to agent-friendly site best practices and emerging protocols like the Universal Commerce Protocol. If an agent completes a task on your site, the accessibility tree stops being a compliance checkbox and becomes an interface. That is a real change, and notably it is a build concern rather than a content one.

If you want help pressure-testing where you actually stand across these surfaces, that is the shape of our SEO and GEO work.

What to actually change