Case study · B2B tech company · October 2026
4.89M Google AI impressions in three months.
That's how often Google's AI showed one tech company in its answers over three months.
Which five things did they do during those three months? Keep reading. The company is under NDA, so we can't name it. But every number on this page is from actual client dashboards.
The screenshots
The last number comes from Semrush, an SEO data tool. The site now has 9,499 search terms on Google's first page, up from 7,046 in January.
What we didn't do
Some AI marketing voices recommend the things below. We don't do them. Either they don't help, or they hurt the brand over time. We build for what's good for the brand in the long run.
- We didn't buy links.
- We didn't send press releases to win AI mentions.
- We didn't add an "llms.txt" file and call it done. Some people treat that file as a shortcut to AI.
- We didn't fake "last updated" dates. We change dates only when the page changes.
- We didn't invent reviews or author bios. We assigned one person to be the public face.
- We didn't hide text or add FAQ markup for questions the page doesn't answer.
- We didn't treat an AI visibility tool as the strategy. A tool shows numbers. Someone still has to act on them.
The five things that mattered most
1. Add what the top results don't already say
Before we write anything, we read the ten pages Google already shows for that topic and note what they all cover. A page still needs those basics, but we aim for about 30% of it to be new: things none of those ten pages say. If we can't add anything new, we skip the topic.
2. Put the answer first
Each section opens with the answer, then explains it. Headings are the questions buyers type into Google. Each paragraph makes one point, so an AI can quote a single paragraph and it still makes sense. Where we can, we give a number instead of a word like "fast" or "powerful".
3. Answer the follow-up questions too
When someone asks an AI a question, it also looks up several related questions in the background. So we list what a buyer would ask next and answer it on the same page, plus a short FAQ built from questions people search for.
4. Check every claim
Every fact needs a source we can point to. Product details get checked against the official spec sheet. Then someone tries to prove each claim wrong before the page goes live.
5. Explain the basic terms
The three pages Google's AI showed most all explain a basic term. Each one links to the product the term is about, so a reader who arrives from an AI answer is one click from the product.
It's never just five things
Behind these five there's a 33-step writing and checking process, brand and legal rules, technical fixes so Google can find and read every page, editing rounds, and tracking for 836 search terms, including whether AI answers mention the site. These five made the biggest difference.
Do ChatGPT, Google and Copilot mention your company?
Message Dom on LinkedIn with your website and the topic you'd want them to mention you for.
Message Dom on LinkedInAuf Deutsch: die ganze Fallstudie und GEO Agentur mit Belegen, kurz zusammengefasst, mit den fünf Dingen und wie du KI-Sichtbarkeit misst.
Data: Google Search Console, Bing Webmaster Tools, Semrush. Client anonymised under NDA. Prepared by Just Digital, justdigital.world.