GEO or SEO? What Google says, and what our own measurement shows

Short answer: According to Google's own documentation, GEO is not a separate discipline. It is part of search work. Google writes "and thus still SEO". For Google's AI Overviews and AI Mode, a page must be indexed and eligible to appear with a snippet. No Google index, no citation in those two features. This is not a universal rule for all AI systems.

German-language comparison sites currently quote between 500 euros and 15,000 euros per month for a GEO package. For the same label. That is what an analysis of those sites, as of July 2026, shows. When the price of a product varies by a factor of thirty, the label does not describe a product.

That is the situation in which marketing managers are asked to open a second budget next to the SEO budget.

So we did not ask vendors. We looked at what the search and AI companies document themselves, what has been measured independently, and what our own website shows.

What Google writes in its own documentation

Google published a guide on working with the generative AI features in Search. The footer dates it 10 July 2026.

The guide names the market terms first and defines them. "AEO" stands for "answer engine optimization", "GEO" for "generative engine optimization". Google then places the work clearly:

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

The reason is given as well. The generative features rest on the same ranking and quality systems as classic Search.

Google gets unusually direct about the tool market:

"Be wary of third-party tools that promise ranking success or claim to use 'internal' Google metrics. No third-party tool has access to our internal ranking or AI systems."

So a dashboard promising a metric from inside Google sells you an estimate. That can still be a useful product. It is simply not a measurement at the source.

The precondition almost nobody mentions

The practically most important sentence sits in a second document, the page on AI features, dated 10 December 2025:

"To be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet."

In plain terms: your page has to be in the Google index and allowed to show a snippet there. Otherwise it does not qualify as a source.

That is not advice. It is an eligibility rule. A page that is not indexed cannot be cited in those features. No service package changes that.

Google also describes a technique called "query fan-out". The system issues several related queries at once and pulls in additional results. That widens the pool of possible sources. It does not touch the eligibility rule.

Four measures Google explicitly calls unnecessary

The guide has a section for this. It is titled "Mythbusting generative AI search: what you don't need to do". Four items from it appear in almost every GEO offer we have seen.

llms.txt and extra machine-readable files. Google writes: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search."

Breaking content into small blocks. Chunking is treated as mandatory across the industry. Google writes: "There's no requirement to break your content into tiny pieces for AI to better understand it."

A special writing style for AI. Google writes: "You don't need to write in a specific way just for generative AI search."

Structured data as a lever for AI answers. Google writes: "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add."

Structured data still makes sense. It feeds the enhanced presentation in the results, not the AI answer.

If you receive an offer listing these four items as deliverables, you now have a basis for the follow-up question.

Our own measurement: the historical snapshot of 14 August 2026

Since 14 August 2026 we run a measurement series on the indexing of our own website. The reason was our move at the time from waterproof-web-wizard.de to waterproof.agency.

The baseline, as of 14 August 2026 (indexing status per Search Console as of 7 August 2026):

In the historical snapshot of 14 August 2026, 182 technical articles were in the sitemap. None had impressions during the measured period. Impressions and indexing are different measures: zero impressions alone does not prove that a page is not indexed.

For the 196 pages reported as not indexed at that time, Google's eligibility rule meant they could not appear as sources in AI Overviews or AI Mode at that indexing status. That alone was not a quality judgement about their content.

A GEO package would not have removed the indexing requirement. The technical investigation had to start one step earlier.

At that measurement point, the migration to waterproof.agency was ongoing. We had implemented technical measures on 13 and 14 August and set a decision point for calendar week 39. That baseline is historical; the following update describes the later status.

Update of 25 September 2026: we are back on our original domains

We wrote this article in the middle of a domain migration. We now know how it ended: we called it off.

What we did. On 26 June 2026, we moved from waterproof-web-wizard.de to waterproof.agency. Every old URL redirected to the matching new page with a 301. On 13 August, we submitted the change of address in Search Console. At the same time, we kept publishing new articles and checked indexing in Search Console week by week.

What came of it. After almost three months, Search Console reported 4 indexed and 624 non-indexed pages for waterproof.agency, as of 21 September 2026. 216 of them were listed as "Crawled, currently not indexed". The new domain built no visibility in that time. And the visibility of the old domain did not noticeably carry over.

Why we went back. We had set ourselves a threshold for calendar week 39: at least 20 indexed pages. We were far below it. So on 24 September 2026, we switched. German content is back on waterproof-web-wizard.de, English content now lives on waterproof-web-wizard.com. waterproof.agency remains our email address and redirects page requests to the matching destination.

What happened next. One day after the switch, we checked 38 newer German blog posts one by one via the Search Console URL Inspection API. Google reports 34 of them as indexed on waterproof-web-wizard.de, all crawled on 24 September. The clearest group are the 21 posts published after the migration: 19 of them are now indexed on .de. None of them had been indexed under waterproof.agency.

One limitation belongs here. The API shows that a page is in the index, not since when. The finding says nothing yet about visitors or enquiries. We will measure impressions again on 1 and 8 October.

What we take from this. Clean redirects and a submitted change of address are the precondition for a migration. They are not a guarantee. Google decides for itself how quickly it transfers signals to a new domain. In our case, almost nothing arrived in three months, although we published regularly. After returning to the established domain, Google reported most posts as indexed just one day later.

This is our single case, not a general rule. Our .de domain had a long history, a new generic domain starts from zero. Anyone planning a domain change should decide beforehand how they will tell after a few weeks whether the move is working. And what they will do if it is not.

For the topic of this article, that means: pages that are not indexed cannot appear as a source in AI Overviews or AI Mode. Three months of migration were three months in which our new posts had no chance there.

Our conclusion: we do not currently recommend a domain migration

According to Google's guidelines, our migration should not have been a problem. Google writes that new generic top-level domains such as .agency are treated like other domains for crawling, indexing and ranking. A move to one is handled like any other site move. In practice, our experience was different.

That is why we currently advise companies against moving to a new domain unless there is a compelling reason. Whether it was the generic ending or the fact that the domain was new and without history, we cannot separate. Both came together. With a new generic domain, we would be particularly cautious today.

Google itself gives a hint, in the same post on the new endings: domain changes can take time in search. It is therefore usually best to choose a domain that fits for the long term from the start. In its guide to site moves, Google says a medium-sized site can take "a few weeks" or longer. For us, it was almost three months without a visible result.

We are not alone in this. In SEO forums on Reddit, site owners report migrations with clean 301 redirects where the new domain is barely indexed months later. One example: a move from .net to .org, with more than 40,000 URLs on "Crawled, currently not indexed" after six months and only one or two indexed pages. The owner writes that a few years ago, such a change took two to four weeks. These are experience reports, not measurements. But the pattern matches ours.

What did not help: Google Ads. From late June to early August, our ads pointed directly to pages on waterproof.agency. That brought around 400 paid clicks to the new domain. It had no visible effect on indexing. That fits Google's own statement: investment in paid search has no impact on organic ranking.

Bing behaved differently. Bing Webmaster Tools lists more than 600 URLs for waterproof.agency as indexed, as of 25 September 2026. Bing discovered many pages on the day of the move and crawled them regularly afterwards. Same domain, same redirects, same content: Bing indexed it, Google did not.

Our assumption, explicitly as an assumption: since more and more content is produced at scale and with AI, Google checks new domains more strictly. Google has a dedicated spam policy against large numbers of pages without added value that mainly aim to influence rankings. Regardless of whether a person or an AI wrote them. Google has not said that new domains are indexed more slowly because of this. We cannot prove it either. But it would explain why so many of our pages stayed on "Crawled, currently not indexed". In the Reddit thread on the .net to .org move, one commenter suspects the same: stricter spam checks make migrations slower today than they used to be. That is no better proven there than it is for us.

If you have to move anyway, decide two things beforehand. First: how will you tell after a few weeks whether the move is arriving? For us, that was 20 indexed pages by calendar week 39. Second: what will you do if it is not? We had kept the old domain and could therefore go back. We would advise everyone to do the same.

Where the other side has a point

This would be too easy if it ended here. There is data supporting a separate view.

Citation patterns overlap only partly with rankings. An analysis of 18,377 matched queries, reported in Search Engine Journal, compared the domains cited by AI systems with Google results. Perplexity's median domain overlap with Google was 25 to 30 percent, ChatGPT's 10 to 15 percent. Gemini varied widely and shared only 160 domains with Google in total. Ranking well in Google does not put you in the AI answer automatically. Limitation: the analysis comes from a vendor of SEO tools, 89 percent of the queries were Perplexity queries, the window covers two months, and the report calls it a snapshot itself.

Editing content does measurably work. The paper "GEO: Generative Engine Optimization" by Aggarwal and colleagues, published at KDD 2024, tested targeted text changes. Quotations, concrete statistics and cited sources raised visibility in generated answers. At the top end by around 40 percent. Limitations the authors state themselves: the benchmark largely ran against a rebuilt pipeline using GPT-3.5 as the answer generator. The paper was submitted in November 2023, before AI Mode and before today's systems. The authors also expect future models with larger context windows to take in more sources, which would reduce the influence of search rankings on the selection.

AI traffic converts well. Karl Wells of the Washington Post told Digiday that visitors from AI platforms spend more time on site. He describes a "four-to-five-times higher subscription conversion rate compared to traditional search". Few visitors can still be worth a lot.

And the retrieval architecture genuinely is new. Michael King of iPullRank argues in the same Digiday piece around "content chunking", the splitting of content into retrievable passages. His point is that every channel needs its own attention.

We take these objections seriously. They show that source selection in AI systems works differently from a ranking. They do not show that this becomes a separately purchased discipline with its own budget.

Jeremy Moser of uSERP puts it most plainly in the same piece:

"If a GEO service does not openly tell you that success in AI visibility is 80 percent good fundamental SEO, they are selling you snake oil."

What has actually been measured in German B2B

For German industry our research found exactly one study with disclosed methodology. It comes from morefire, published on 14 July 2026 by Robin Heintze.

The setup: 62 German B2B industrial companies, 12,250 evaluated AI answers, four systems. Plus ten neutral questions per company and five runs per question and system.

Two results matter in practice.

First, the share of answers. On average a company appears in 6.6 percent of the answers in its own category. The best value in the field is 17.2 percent.

Second, and this is the real finding: of 118,099 citations, only 5,955 pointed to the company's own domain. That is 5 percent. The remaining 95 percent came from third-party sources, among them industry directories and Wikipedia.

So what decides your visibility is mostly what other sites say about you.

Two cautions apply. The study comes from a vendor selling a tool for this. And it discloses a weakness that affects the whole measurement market. 46 percent of the brands did not appear in all five identical runs. Same question, same system, different answer. Gemini performed a live web search in 0 percent of cases, Perplexity in 99.7 percent.

Anyone buying a visibility metric for AI answers is buying a number with substantial spread. That does not make measuring pointless. It means running repetitions and not over-reading single values.

How large this channel really is today

Two numbers help against both kinds of exaggeration.

SE Ranking analysed Google Analytics data from 101,574 websites. ChatGPT's share of referral traffic was 0.23 percent in April 2026 and 0.32 percent in May. That is a 36.7 percent rise in one month and the highest value measured so far. The authors add themselves that this traffic makes up "only a tiny portion of total website traffic".

At the same time click behaviour in Search is shifting. Pew Research Center has the cleanest data set here. 900 adults in the United States were measured through their browser. That produced 68,879 searches in March 2025, of which 12,593 showed an AI summary. With an AI summary, users clicked a classic result in 8 percent of visits. Without one, in 15 percent.

Taken together: direct traffic from AI systems is small today and growing fast. The loss of clicks in classic search is already noticeable. Taking both seriously at once is the uncomfortable but honest reading.

What the job market does

One signal we found along the way, and it convinces us more than any position paper.

Companies posting roles and projects in the German-speaking market look for a "SEO/GEO Manager" or "SEO & GEO Manager". As one role. That holds for agency networks as much as for specialist service providers. The combined title even shows up at trainee level.

Our research found no posting looking for GEO in isolation, without an SEO scope.

Companies paying salaries for this have settled the question in practice. They fill one position, not two.

The evidence at a glance

Three of the five come from companies selling something in the same market. That does not void the numbers. It belongs in the picture.

What we take from this for the work

Three stages, in this order. The order is the actual advice.

Stage 1: establish eligibility. Check how many of your pages are actually in the index. The number sits in Google Search Console under Indexing. If a large figure shows under "Crawled, currently not indexed", that is your topic. Everything else stays theory until then. This is the core of our SEO consulting for B2B.

Stage 2: make the text citable. This is where the research does deliver something. Your own figures, named sources and clearly bounded passages raise the chance that a system picks up your section. A paragraph that answers a question completely is easier to cite than a text building towards a conclusion. How we plan articles with substance of their own is described on the service page.

Stage 3: take third-party sources seriously. If 95 percent of citations in your category do not come from your domain, the rest of the web has a say. That means industry directories, trade media, associations and databases where your company is listed correctly and currently. It explicitly does not mean bought mentions. Google warns against building artificial mentions in the same guide.

The difference is simple. A directory entry with correct data is maintenance. A purchased swarm of mentions is a risk.

Check it on your own site

Four questions you need nobody for.

1. How many of your pages are indexed, and how many are not? If you do not know the number, that is the first task.
2. Do your most important articles contain a figure, a measurement or a case from your own operation? If not, the text is replaceable.
3. Are your entries in industry directories current? Company name, address, services.
4. If someone sends you a GEO offer: which line item in it is not search work? If none is left, it is a relabelling.

What is at stake, soberly

Ignoring the topic will not cost you your business overnight. Direct traffic from AI systems currently sits in fractions of a percent.

What you lose is less visible. A measurement series only becomes meaningful over time. If you record no baseline today, you cannot say in a year whether anything changed. The value comes from the history, and history cannot be created retroactively.

Releasing a second budget while your own pages are not in the index pays for the roof before the foundation.

Next step

You know our numbers now. You do not know yours yet.

If you want to know how many of your pages are citable at all, we will look at it together. We go through your indexing status, check your most important pages for substance of their own, and tell you where the bottleneck sits.

The first call takes 15 minutes and costs nothing.

• Google Search Central, AI optimization guide, 2026-07-10. developers.google.com
• Google Search Central, AI Features and Your Website, 2025-12-10. developers.google.com
• Google Search Central, Creating helpful, reliable, people-first content, 2025-12-10. developers.google.com
• Pew Research Center, 2025-07-22. pewresearch.org
• Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024. arxiv.org
• morefire, AI visibility study, 62 German B2B industrial companies, 2026-07-14. more-fire.com
• Digiday, GEO hype busted, Sara Guaglione, 2026-03-10. digiday.com
• SE Ranking, ChatGPT referral traffic, Yulia Deda, 2026-07-09. seranking.com
• Search Engine Journal on the Search Atlas analysis, 2025-11-19. searchenginejournal.com
• OpenAI, crawler documentation. developers.openai.com
• Anthropic, crawler documentation, 2026-04-07. support.claude.com
• Perplexity, crawler documentation. docs.perplexity.ai
• SEYBOLD, What does GEO cost? Analysis of German-language comparison sites, as of July 2026. seybold.de