One question comes up more and more often — and it comes from companies that have done their homework: We rank near the top on Google. So why doesn't ChatGPT mention us?
Our own Search Console data now carries queries nobody would have typed two years ago. Whole sentences instead of keywords. One of them, in substance: "ChatGPT describes our company using outdated product data — who in Switzerland fixes that?"
This is no longer a fringe case. It is a second results surface with different rules, and on that surface a number-one Google position guarantees nothing.
Ranking and citation are two different outcomes
The most important sentence first, because it orders the whole debate:
A good Google ranking is still necessary. It is simply no longer sufficient.
That is not an opinion. In March 2026 Ahrefs analysed roughly 863,000 search result pages and four million URLs cited in AI Overviews. The finding: only about 38 percent of cited pages also rank in Google's top 10. In July 2025 it was 76 percent.
One honest caveat belongs with that number: Ahrefs attributes part of the drop to improvements in its own parsing, not solely to changed behaviour at Google. The direction still holds.
For you it means this: the set of pages an AI answer draws from now only partially overlaps with the set your SEO is aimed at.
Why search is splitting in two
Search engine optimisation used to rest on four things: keywords, backlinks, meta data and a technically clean website.
All four still apply. What has been added is a second layer, where systems such as ChatGPT, Google AI Overviews, Google AI Mode, Bing Copilot and Perplexity decide which source they name in their answer.
These systems judge differently. They read context, intent, semantic relationships, entities, freshness — and above all whether a statement shows up consistently in more than one place.
Three terms have settled around that second layer:
- GEO — Generative Engine Optimization: preparing content for generative AI systems.
- LLMO — Large Language Model Optimization: preparing content for language models such as ChatGPT.
- AEO — Answer Engine Optimization: optimising for the direct answer rather than the click.
We took those three apart in AI SEO explained. This article answers the other question: which parts of it are measured, and which are merely claimed?
What is proven, and what is not
This is where it gets uncomfortable, because much of what is currently sold as "AI SEO" does not survive scrutiny.
The most thorough test so far comes from research: C-SEO Bench, published at NeurIPS 2025, examined 54 combinations of method and domain. Three were significantly positive. None of them in question answering. A review of 45 studies from July 2026 reaches the same conclusion: none of the techniques reviewed shows a stable, cross-platform effect.
One piece of advice is especially persistent, repeated in guide after guide — that structured data increases your chance of being cited by AI. Ahrefs tested it properly in May 2026: 1,885 pages with newly added schema markup against roughly 4,000 control pages. The result for AI Overviews was minus 4.6 percent — significantly negative. For AI Mode and ChatGPT the effect was indistinguishable from zero.
Google itself is unambiguous: there are no additional requirements and no special optimisations needed to appear in AI Overviews or AI Mode.
Schema markup is still the right thing to do. It makes your content machine-readable for classic search and carries rich results. It is simply not an AI lever — and selling it as one means working towards an effect that does not exist.
The four things that measurably hold
Strip out what cannot be evidenced and the list gets shorter. It also gets reliable.
1. The crawlers have to get in at all. OpenAI states it plainly: block “OAI-SearchBot” and you do not appear in ChatGPT Search. That applies to robots.txt as much as to firewall and CDN rules nobody remembers setting. It is the cheapest check in the field and the one most often skipped.
2. The page has to be indexable and snippet-eligible. It is the only condition Google names. And it has an unpleasant flip side: a “nosnippet” or “max-snippet:0” removes a page from AI Overviews and AI Mode — without hurting rankings and without any tool raising an alarm.
3. Classic ranking. Necessary, not sufficient — see above. And peer-reviewed as the more effective route.
4. Topical coverage. The most reproducible lever in the entire review. Covering a subject completely beats any individual optimisation technique.
There is a fifth point, which is not a lever but an observation about where citations come from. Muck Rack analysed more than 25 million links out of ChatGPT, Claude and Gemini: 84 percent of citations come from editorial sources, 0.3 percent from paid ones. Stable between 82 and 89 percent across three surveys since July 2025.
Be careful turning that figure into an action. What is proven is where citations originate — not that a single media placement puts your brand into a particular answer. What it does say clearly: budget spent on bought placements is badly spent here.
Technical SEO remains the foundation
None of this replaces the basics. Without clean technical ground, neither Google nor a language model processes your content reliably:
- fast load times and passing Core Web Vitals
- mobile optimisation
- HTTPS without mixed content
- readable URL structures
- a maintained XML sitemap
- internal linking that leaves no page stranded
It is unglamorous, and that is exactly why it works. What that layer looks like in practice is in our article on technical local SEO.
Why you have to measure this rather than hope
Here lies the biggest departure from classic SEO, and the most underestimated one.
In January 2026 SparkToro analysed nearly 3,000 runs across ChatGPT, Claude and Google. The same prompt returned the same list of brands in under one percent of cases. The same order in under one in a thousand.
That has a hard practical consequence: a single AI answer is not a measurement. Asking ChatGPT once, taking a screenshot and drawing a conclusion from it measures noise. It only becomes reliable with several runs, fixed questions and a repeated point in time.
Choosing the right metric matters just as much. Similarweb measures a citation rate of under four percent for professional services — our sector, and the weakest of all sectors. That figure is the share of answers carrying any source attribution at all. Measure yourself against citation rate and you are measuring against a ceiling you cannot break through.
The mention rate — how often your brand is named, with or without a source link — is the more meaningful figure. A mention without a link counts fully there, and a recommendation is exactly that.
What this means for Swiss SMEs in practice
The good news: in Switzerland this field is still largely unoccupied. There is almost no public data on how Swiss providers appear in AI answers — and whoever measures it themselves holds something no model can invent and no competitor can copy.
A sequence that works:
- Check access. Do the AI crawlers reach your site? One hour of work, once.
- Check snippet eligibility. Is there a “nosnippet” anywhere you did not set deliberately?
- Measure the starting point. Fixed questions, several runs, written down. Without that zero point, every later statement about progress is a guess.
- Cover the topic completely instead of working single terms.
- Become visible where editorial coverage happens — trade media, associations, real articles.
What you will not find in that list: a technique that promises you a placement. There isn't one, and in June 2026 Google published a page of its own aimed at exactly those promises.
We build the framework at ALPENIQ Growth and the technical side at ALPENIQ SEO.
The honest conclusion
Search has not been replaced. It has gained a second output channel that favours different sources and behaves slightly differently on every query.
Anyone responding to that today is well served by entering the field soberly: four levers are evidenced, one of the most frequently named is refuted, and without your own measurement all of it is assertion.
That is less than most providers promise. It is also the part that holds.
Common questions
Is a good Google ranking still enough?
No, but it remains the precondition. Ahrefs measured in March 2026 that only around 38 percent of pages cited in AI Overviews also rank in Google's top 10 — down from 76 percent in July 2025. Ranking is necessary but guarantees no mention.
Does structured data help you get cited in AI answers?
On current evidence, no. Ahrefs compared 1,885 pages carrying new schema markup against roughly 4,000 control pages and measured a significantly negative effect of 4.6 percent for AI Overviews. Schema remains right for classic search and rich results — it is just not an AI lever.
How do I measure whether my company appears in AI answers?
With fixed questions, several runs per question and a repeated point in time. SparkToro showed that the same prompt returns the same list of brands in under one percent of cases, which is why a single answer is not a measurement but a sample of one.
What is the difference between citation rate and mention rate?
Citation rate counts only answers that carry a source attribution. For professional services Similarweb puts it below four percent — sector-wide, not company-specific. Mention rate counts every naming of your brand, link or not, and is therefore the more robust figure.
Is it worth buying placements in online media?
For AI visibility, hardly. Muck Rack analysed more than 25 million citations: 84 percent come from editorial sources, 0.3 percent from paid ones. What that proves is where citations originate — not that a single placement triggers a mention.
What is the most common technical mistake here?
An exclusion nobody remembers. If a firewall, CDN or robots.txt rule blocks “OAI-SearchBot”, the page does not appear in ChatGPT Search — OpenAI says so explicitly. A forgotten “nosnippet” does the same thing: it removes the page from AI Overviews without touching rankings.