From SEO to GEO: what large companies really change
Since May 2026, Google has documented that GEO, from its point of view, is still SEO: its generative features rely on its usual ranking systems. What large companies changed is therefore not a discipline. It is their content standard, their measurement and the person accountable for the answer.
- On 15 May 2026, Google published its first official documentation on the subject and writes there that optimizing for its generative features is SEO.
- No proprietary file or markup is required. Google explicitly sets aside the llms.txt file, content chunking and Markdown versions of pages.
- Since 3 June 2026, Search Console has offered a dedicated report on impressions in AI summaries and conversational mode. It gives impressions, not clicks.
- What sets visible companies apart is not a tool. It is content no one else can produce, what Google calls non-commodity content.
- The budget to create is not a second content budget. It is a measurement line item and a named responsibility.
On this page
Optimizing for generated answers
GEO, for generative engine optimization, is the work of getting a company to appear in the answers produced by generative systems. Google states in its documentation that, from its point of view, this work is SEO, because its generative features rest on its usual ranking and quality systems. The useful distinction is therefore not about discipline: it is about the surfaces targeted, which extend beyond Google, and about measurement, which follows rules other than traffic.
What Google finally put in writing
For two years, this subject sold in Quebec without anyone being able to rely on anything more than interpretations. That period is over. On 15 May 2026, Google published documentation aimed at site owners, under a new section devoted to the fundamentals of generative AI, and its position there is explicit. Good SEO practices remain valid because the generative features of Search rest on its ranking and quality systems.
Google goes further and names the market's terms. It acknowledges the acronyms AEO and GEO, specifies that, from the point of view of Search, optimizing for generative AI amounts to working on the search experience, and points companies to its advice on evaluating third-party SEO services. In other words, Google treats the subject as a question of vendor quality.
The documentation also explicitly sets aside a series of practices sold as prerequisites. The llms.txt file, which Google treats like any text file with no special handling, chunking content into fragments ahead of time, rewriting aimed at every long variant of a term, and producing Markdown versions or special markup. For a management team, each of these line items is a cost a vendor may have billed and that no result will ever justify.
This clarification does not say the subject is empty. It says the value is not found in a proprietary technique. It is found in content competitors cannot reproduce and in a measurement almost no one keeps. These are two projects that are less easy to sell and far more profitable.
What to settle before opening an AI visibility budget
The financial risk on this subject is not doing nothing. It is paying twice: an SEO budget and an AI visibility budget, for work Google describes as the same. The question to settle is not whether the subject matters, but whether it justifies a separate line item.
- Does what our vendor bills under GEO differ from what it already bills as SEO, and how exactly?
- Do we have data, prices or cases no one else can publish, and are they published?
- Who, on our side, is responsible for what the AI answers when we are named?
- On which questions do we want to be the cited source, and are they the ones that bring in customers?
- If no citation is recorded within nine months, what do we stop?
The useful answer draws on Google's documentation, clearly separates measurable surfaces from the rest, and talks about proprietary content. A hollow answer sells an in-house method, bills for a particular file or markup, or promises traffic from AI without being able to show where the figure comes from.
Deciding whether this work deserves a budget separate from your SEO budget is a call that is settled on your real content inventory. A 90-minute consultation settles it, with a written summary your team can execute.
What large companies actually changed
The organizations that appear regularly in generated answers did not create a dedicated department. They changed three things, and none of them is a tool.
They raised the content standard
Google stresses unique, non-commodity content. A summary of what already exists feeds the answer without ever getting its source named. What gets cited is what no other company owns: your figures, your real timelines, your prices, your documented cases. It is the only investment in this work that produces an asset, and it is also the most uncomfortable, because it forces you to publish what competitors cannot copy.
They named someone accountable for the answer
In most companies, no one is tasked with checking what an AI answers when a potential customer asks who to turn to in the sector. The work is to record the answer against a fixed list of questions, note who is cited, and correct what is wrong. It is one hour a month, and it is the difference between steering and hoping.
They separated traffic measurement from presence measurement
Being cited produces very few visits. Confusing the two leads you to cut work that was building awareness. The companies that go the distance track two separate dashboards, with two decision logics. The content structure that makes a page reusable is covered in structuring content to be quoted.
The line between in-house and delegated shifts here compared with classic SEO. What stays in-house is heavier than elsewhere: the proprietary data, the real prices, the timelines actually met, the lived cases with their figures, and the permission to publish them. It is that permission that blocks things most often, and no vendor can get around it. What can be delegated: the list of questions to watch, the monthly recording of answers, the restructuring of existing content, source checking and tracking in Search Console. A company that agrees to publish three data points its competitors keep to themselves earns more citations than a company that produces twenty cautious articles.
Measurement exists at last, and only halfway
On 3 June 2026, Google launched a dedicated report in Search Console for the generative features of Search. It covers AI summaries and conversational mode, with a separate report for Discover. The rollout happens in stages and not all properties have access to it yet.
What the report gives: impressions, broken down by page, country, device and date, with export. What it does not give: clicks, and therefore no way to tie a commercial value to a presence. Google also specifies that data from Search Labs experiments is excluded, and that the usual Search Console limits, including the thousand-row cap, apply.
| What you measure | Where | Reliability | Cadence |
|---|---|---|---|
| Impressions in AI summaries and conversational mode | Search Console, dedicated report | Good, official source | Monthly |
| Most reused pages | Search Console, pages dimension | Good | Monthly |
| Presence on surfaces beyond Google | Manual recording or tracking tool | Moderate, sampled | Monthly |
| Share of voice against cited competitors | Recording or tracking tool | Moderate | Quarterly |
| Clicks from generated answers | Nowhere at Google to date | None | Do not track |
| Sales attributed to a citation | No reliable source in the industry | None | Do not track |
The last two rows are the ones that tell a serious vendor from another. No one today can honestly link a citation in a generated answer to a closed sale. A provider who promises it will have to produce a figure at the end of the engagement, and that figure will be fabricated.
Search Console also offers a control that excludes a site from the generative features of Search. This setting is a management decision, not a technical one: it affects the brand's visibility and there is no reliable measure of the traffic you give up. Deciding it without an explicit trade-off is a misplaced commercial risk.
The surfaces that escape Google
All the documentation cited so far talks only about Google. That is the limit of the "GEO is SEO" reasoning: it is accurate for the generative features of Google Search, and it says nothing about the conversational assistants your customers use elsewhere.
On these surfaces, there is no comparable official documentation, no measurement report provided by the platform, and no guarantee of stability from one month to the next. Measurement there is a recording, manual or tooled, against a list of questions fixed in advance. It is useful, but it is sampled: it indicates a trend, never a volume.
The budget consequence is direct. The underlying work is common to both worlds, which proves Google right. The measurement work is not, and that is where the only honestly separate line item sits. The general framework for SEO is laid out in the complete guide to search, and the effect of summaries on traffic in Google AI summaries.
What you should not pay for
A proprietary method. Google writes that its generative features rest on its usual systems. There is therefore no back door to buy. A vendor who sells privileged access to generated answers is selling something that does not exist, at the price of an innovation.
A second content inventory. Producing parallel pages meant for AI amounts to paying twice for the same subject and creating internal competition among your own pages. Reworking existing content costs a fraction of the price and produces more.
A report that counts mentions without qualifying them. Being named in an answer to a question no one asks before buying is worth nothing. The list of monitored questions must come from your salespeople and your customer service, not from a keyword tool. The topic of content produced with assistance is covered in what AI changed about content, and the writing rules in what changes and what does not in writing.
Google has just written that the back door does not exist. What remains is what no one wants to sell: publishing what your competitors cannot publish.
Falia analysis gridCheck before renewing an AI visibility budget
The indicators that change a decision are detailed in marketing indicators, and the execution on the GEO and AEO page.
This point sits inside the framework described in who measures what, and who measures nothing.
Visibility in answer engines is at the heart of the Strengthen your visibility in AI answers goal.
Already running a marketing team? See how we plug in as reinforcement on optimizing for answer engines.
Frequently asked questions about the move from SEO to GEO
Is GEO a discipline distinct from SEO?
Not from Google's point of view. Its documentation published in May 2026 states that its generative features rest on its usual ranking and quality systems, and that optimizing for them therefore amounts to doing SEO. The useful distinction is about the surfaces beyond Google and about measurement, not about method.
Do you need an llms.txt file?
No. Google states that its crawler can discover this file but treats it like any text file, with no special handling and no privileged indexing path. Other crawlers may make use of it, but no billing for this work is justified by visibility in Google.
Do you need to chunk your content into fragments for AI?
No. Google specifies that its systems understand pages covering several subjects and can extract the relevant passage without the author having fragmented the text in advance. What helps is still a clear structure, not artificial chunking.
Can you measure your presence in generated answers?
In part. Since June 2026, Search Console has offered a dedicated report on impressions in AI summaries and conversational mode, rolling out gradually. It gives impressions by page, country, device and date, but no clicks. Surfaces beyond Google are measured by recording, with lower reliability.
Can you attribute sales to citations in an AI?
No, not reliably, and no player in the industry can do it today. What is measured is presence and its evolution. A vendor who promises attribution of sales from generated answers is announcing a certainty that does not exist.
What should you change first?
Publish what no one else can publish. Google emphasizes unique, non-commodity content, and it is the only factor that durably sets a cited company apart from a company that feeds the answer without being named. Your figures, your real timelines, your documented cases.
- Google Search Central, Optimizing your website for generative AI features on Google Search, official documentation published on 15 May 2026, accessed July 2026.
- Google Search Central, A new resource for optimizing for generative AI in Search, 15 May 2026.
- Google Search Console, Help Center, Generative AI performance report (Search), official documentation, accessed July 2026.
- Google Search Central, Introducing Search Generative AI performance reports in Search Console, 3 June 2026.
Geneviève puts the strategy for your engagement into action. She leads all our web development projects: Shopify, WordPress and the new ways of building a site with AI. She manages our team of developers and translates your business needs into technical language. She runs your organic search (SEO), your visibility in AI answers (GEO) and your site's conversion rate optimization (CRO). Her work is at the heart of three goals: Attract customers with SEO and AI, Improve your site's conversion, and Strengthen your visibility in AI answers. With Gabriel, she also builds the landing pages for your advertising campaigns. She writes mainly about SEO, AI visibility and web design.
About Falia →
SEO
Paid advertising
Conversion
AI visibility
Web design