KnowledgeFundamentals
ChatGPT Optimization: How to Get Recommended
ChatGPT optimization is the systematic work of getting ChatGPT to name your company when a potential customer asks it for a provider. It covers three areas: the sources ChatGPT draws on for your topic, your own content, from which the system takes citable statements, and the correction of false information about your company. There is no fixed placement to win. What you can measure is the share of answers in which your company is recommended.
Why provider selection now happens inside ChatGPT
The shortlist a buyer contacts is increasingly decided before they open a website. According to a Comparis survey conducted by Innofact in March 2026 among 1,035 people, more than three quarters of adults in Switzerland (76%) use AI tools in their daily lives; a year earlier the figure was 62.4%.
In our daily measurements, an AI answer names typically three to five providers.
You are in or you are out. There is nothing in between.
What ChatGPT does differently from a search engine
ChatGPT keeps no ranking, searches the web live for current questions, and does not give every user the same answer. These three properties are why search engine optimization alone falls short.
There is no ranking
ChatGPT generates every answer from scratch. The same question returns a different set of providers on the next run. Whether a ranking in ChatGPT exists at all is covered in its own article.
ChatGPT researches live
For current questions, ChatGPT searches the web and reads individual pages instead of relying on training knowledge alone (OpenAI documentation). Your website supplies the material the answer is built from.
Every user gets a different result
ChatGPT can remember earlier details across separate chats and factors in context such as history and preferences (OpenAI documentation). Two prospects asking the same question therefore see different recommendations.
One question becomes several
Almost nobody types the fully formed question that is being measured. The user writes «I am looking for a fiduciary firm», the system enriches the request with context in the background and turns it into several more specific queries. The term for this is fan-out.
What you optimize for is therefore not a keyword but the set of questions your buyers actually ask. How a set of 50 real buyer questions is put together shows where that set comes from.
The three levers: sources, content, corrections
ChatGPT optimization works on three fronts. All three act together, because the system checks the wider picture across several sources and a claim you only make on your own website does not carry it.
| Lever | What it covers | Typical work |
|---|---|---|
| Source authority | Presence on the sources AI systems demonstrably cite for your topic | Directories, association lists, trade pages, profiles, partner sites |
| Answer integrity | Preparing your content so machines understand what you offer and who it is for | Pages that answer buyer questions directly; definitions, comparisons, data; clean structure and crawlability |
| Corrected facts | Setting right what is wrong or outdated about your company | Tracing a statement back to its original source and correcting it there |
The third lever is the one most often underestimated. A false statement about your company almost always has one concrete source, and that source can be found.
What does not work
Tricks. The system wants to deliver the best answer, and relevance, clarity and currency decide which providers it names. Three approaches that get tried regularly and go nowhere:
Keyword density and FAQ blocks without substance
A question-and-answer structure only helps when the answer says something that is not available anywhere else.
Calling yourself the market leader
Saying so on your own website does not get you recommended. The system cross-checks the claim against independent sources.
A tool on its own
A tool shows you that you are not being mentioned. It does not change it.
What does work is work on the places the answer is built from. A study by Aggarwal et al. (KDD 2024) tested targeted content adjustments and measured, for the most effective variants, a visibility gain of up to 40 percent in generated answers. That is a ceiling under laboratory conditions, not an expected average.
How you measure progress
By the recommendation rate: the share of repeated identical queries in which your company is recommended. A single query says almost nothing, because every answer is generated from scratch.
Three conditions make the number reliable: the same questions over weeks, queries run without login and without memory so the measurement does not influence itself, and three to five repetitions per question. The full measurement protocol is set out in how AI visibility is measured reliably. For a first read of your own, the shortened self-test in check your AI visibility yourself is enough; the free AI visibility analysis delivers the same snapshot in structured form, by email.
For a Swiss SME in a contested B2B field, 30 to 40 percent is a strong result. Treat it as a reference point, not a commitment: dozens of providers compete for the few slots in each answer.
How long it takes before anything moves
The first weeks are setup, movement shows later. The timeline below describes the mechanism, not a promise.
Week 1: where you stand today
The baseline documents which buyer questions name you now, which competitors appear, and which sources support the answers.
By day 30: first measures live
The fast levers come first: crawlability, clear answer passages, gaps on sources the systems already trust.
Day 60 to 90: first reliable movement
Still volatile, but readable as a trend.
Month 3 to 6: durable authority
Topical depth and consistent presence across several independent sources. For a new brand or a new domain it is 6 to 12 months or longer.
Anyone who names you a fixed date («in ChatGPT within 30 days») is selling you something. If nothing shows on the fast live systems by day 90, the cause is usually crawlability, unclear company data or weak third-party sources, not a shortage of patience.
Who this is worth the effort for
Swiss SMEs whose customers research and compare before they buy: a B2B focus, services that need explaining, a sale decided on trust.
What decides it is the value of a new customer. Even a small operation pays for itself when a single won contract covers several months of the work. For larger, well-known companies the benefit shifts: there the question is less whether the systems name you than what they claim about you.
One objection comes up often: wait for the website relaunch first. The relaunch is the cheapest moment for this. You build structure and wording correctly once, and the measurement runs from today.
ChatGPT optimization with Carigiet GEO
At Carigiet GEO, visibility in AI systems is an ongoing monthly retainer: strategy, content and implementation. You supply subject-matter input and approvals. We run 50 buyer questions per service area every day across six AI systems: ChatGPT, Claude, Perplexity, Gemini, Copilot and Google AI Overviews. That is 300 runs per day, evaluated as a weekly comparison against the same set of competitors. On top of that, a weekly exchange of 30 minutes.
We set up onboarding and the baseline, maintain the questions, measure daily, produce the content and publish it, build sources and directory entries, and correct false statements. From CHF 2,800 per month plus VAT, no setup fee, no minimum term, cancellable with a 30-day notice period to the end of the following month. Each additional service area with its own set of competitors costs CHF 1,500 per month. Everything produced belongs to you, including after the contract ends. What the retainer covers in detail is set out in what a GEO retainer includes.
The professional grounding is 25 years of software development. Generative Engine Optimization is a young discipline in which nobody can claim long experience; whoever understands how the systems produce their answers can influence them.
