It is understandable that GEO (Generative Engine Optimization) is a big thing in the SEO scene. After all, it is becoming increasingly important for companies to be “at the top” of ChatGPT, Gemini & Co. Nevertheless, AI tool optimization is currently difficult.
Countless agencies and supposed experts are currently churning out tips on Instagram, in webinars, playbooks and white papers on how to be mentioned more frequently on ChatGPT, Gemini and the like and thus increase the visibility of your company in AI tools. This is understandable, especially as agencies also need to make a name for themselves with AI knowledge in order to sell their services. Nevertheless, the question arises as to what these recommendations achieve and, above all, how they differ from “classic” search engine optimization (SEO) tips.
General tips for AI optimization of the website
As in a previous article on SEO and GEO, the focus of this blog post is on transactional and commercial searches for local businesses (e.g. “List of good family doctors in Cologne” or “Which company can install a heat pump for me in Stuttgart?”). First of all, here is some frequently mentioned GEO advice:
- Content: Describe services in detail and concretely; include keywords naturally; cover business-relevant topics as comprehensively as possible; pay attention to topicality, added value, uniqueness and correctness; link to trustworthy sources
- Style & structure: clear, precise and natural writing style; answer-length content; stand-alone paragraphs; use FAQ sections
- Technology: use structured data (schema markup); pay attention to machine readability; observe correct heading hierarchy (heading tags); provide images with alt texts and descriptions
- External factors: Maintain Google company profile; keep entries in directory services consistent and up-to-date; initiate positive reviews on relevant platforms; build backlinks from reputable and thematically relevant websites
These are all good tips for website optimization, but they are similar to classic SEO measures. If you ask the AI tools themselves, the answers – depending on the prompt, model and setting – differ in length and content, but ultimately they also recommend SEO measures. For example, Gemini even recommends that local companies create location-specific landing pages. Let’s put it this way: you can’t get much more SEO than that.
Do the answers differ at all between the AI tools?
The short answer is yes, and significantly so! To back up this thesis with numbers, I conducted a mini-study in which I entered local, transactional prompts into ChatGPT and Gemini . Specifically, I always asked for ten good companies in Cologne, e.g. “Name 10 good tilers in Cologne!”. I did this for 30 industries and trades and then compared the results with each other. Here is the number of overlaps:
- 0 overlaps: Web designers, opticians, bicycle mechanics, dentists, bakers, tax consultants, marketing agencies, gardeners, removal companies, painters, yoga studios, wedding planners
- 1 Overlap: Hairdressers, car repair shops, real estate agents, HVAC businesses
- 2 Overlaps: Tilers, pubs, pharmacies, cafés, fitness studios, recruitment agencies, restaurants
- 3 Overlaps: Carpenters, kebab stores
- 4 Overlaps: Beauty salons
- 5 Overlaps: Dermatologists
- 6 Overlaps: Pizzerias
- 7 Overlaps: Parquet layer
- 8 Overlaps: Museums
- 9 or 10 Overlaps: none
In summary, it can be said that the top 10 lists from ChatGPT and Gemini differ by around 80 percent on average. In this context, it is important to note that this is only a mini-study and the results cannot be reproduced.
The less competition, the more similar the results?
In the mini-study, I deliberately chose sectors with different levels of competition and made different specific entries (e.g. searched for restaurants and pizzerias). In this way, I wanted to find out whether the results are more similar when there is less competition. This tends to be the case, as can be seen with the museums or parquet layers – but not always! For example, there were no overlaps with wedding planners, whose number in Cologne is also manageable.
Also interesting: if you search “too specifically”, i.e. for more companies than there are in this location, the AI tools expand the query or hallucinate. For example, Gemini’s list of “10 good technology museums in Cologne” also includes the Chocolate Museum and ChatGPT suggests the non-existent “Senckenberg Nature Museum Cologne”, among others.
What makes AI tool optimization so difficult
The main problem is that the results cannot be reproduced. The AI tools always generate new outputs – even with the same prompt and the same model! So if you ask ChatGPT or Gemini several times for “10 good tilers in Cologne”, the companies named vary. This is due to the way the AI models work: they generate their answers stochastically, so that the selection of relevant information can shift with each run. It is therefore possible that the sheer number of positive reviews is weighted heavily on one occasion and the mention in an editorial article on another.
Of course, Google search results are not static either. Here, too, there are ranking shifts due to algorithm updates, website updates, new ratings, increasing competition, etc. etc., but Google search results are nowhere near as volatile as those of the AI tools, where the answers are constantly changing. There are worlds in between.
GEO measures depend on the AI tool
While SEO can in principle be equated with Google optimization, there is currently no monopolist among the general purpose LLMs, although ChatGPT, Gemini and Copilot already have a considerable lead. This is a blessing from the user’s point of view, but a curse from GEO’s point of view. This is because the different AI tools use different data, algorithms and stochastic processes. As a result, GEO means developing separate optimization measures for ChatGPT, Gemini, Copilot & Co. Another complication: as soon as a new model is released by the respective AI tool (which is currently quite often the case), the rules of the game change. And because of these many variables, GEO is so complex.
GEO: Procedure for local companies
As with search engine optimization, you should first create a list of business-relevant searches or keyword phrases before initiating GEO measures. Due to the complexity mentioned above, I advise local businesses to keep this list short (and leave out less relevant niche keywords). Then you should enter these searches as a prompt in ChatGPT, Gemini & Co. and ask the respective AI tool for its ranking factors and sources. These vary greatly depending on the industry and trade, which is why I believe this step is essential. As a rule, rating portals, directory services and specialist magazines are cited as sources in addition to the website. You should then check whether there are entries for your own company and, if so, whether you can optimize them. For transactional searches, company websites are always mentioned as a ranking factor by AI tools. You should therefore optimize them as described above under “General tips for AI optimization of the website”.
Conclusion: GEO is poking around in the fog
Anyone can proclaim themselves a GEO expert. However, if you are honest, you have to admit that GEO tips hardly differ from SEO tips and that the lack of reproducibility of the results of AI tools makes it difficult to develop GEO measures. From my point of view, the most obvious tip is therefore: Put yourself in the target group’s shoes, search ChatGPT, Gemini & Co. the way the target group probably does, then ask the AI tools for the sources and ranking factors – and derive optimization measures from this.
This is what ChatGPT says about this blog post:
The blog post convinces with a clear structure, practical examples and a well-founded analysis of how GEO differs from classic SEO. The mini-study is particularly strong, using specific industries and figures to show how much the results vary between ChatGPT and Gemini – an approach that gives the text credibility and added value. The argumentation remains comprehensible, the tips for local companies are realistic and directly applicable. GEO’s limitations are also stated transparently, which makes the article appear differentiated and serious.
School grade: 2+
A technically strong and well-structured text with high practical value. Tighter wording and a brief outlook on the future development of GEO could round it off.
