AI systems don't read reviews as a star rating, but as text: which service is mentioned, which location, which trait keeps repeating. This produces statements like "praised for short-notice appointments", which then feed into recommendations. Businesses that reply to reviews and have gathered them across several platforms provide more consistent evidence. No provider documents how heavily a model weighs reviews.
When ChatGPT recommends a business, the recommendation is usually backed by a sentence like "known for reliable scheduling" or "customers praise the advice". The model doesn't invent sentences like these. They come from text that's published somewhere publicly, and in many cases that's reviews. This page explains how a review becomes evidence, why your replies count towards it, and what's different in Switzerland.
From star to sentence: how reviews become evidence
A language model doesn't calculate an average. It reads text and draws statements about an entity, in other words, about your company. It reads three things from a review:
- The service. "Installed our kitchen" tells the model that this business builds kitchens. "Great work!" says nothing at all.
- The location. "In Amriswil" or "here in Thurgau" links the company to a region, even if the website only mentions this in the legal notice.
- The trait. "Arrived the same day", "stuck to the quote", "included the children": whatever repeats across many reviews becomes a feature.
An example: a plumbing business has 60 Google reviews. In 18 of them, something along the lines of "arrived quickly" appears. If someone asks ChatGPT for an emergency plumber in that region, there's a good chance this business gets mentioned with exactly that trait. Whether the model reads the reviews directly or a summary of the business profile isn't documented. The result is the same.
This gives us the most important rule: don't ask customers for "a review", ask for two sentences about what was done and what stood out to them. That's the kind of review that tells an AI something.
Why your replies to reviews count too
The business's reply sits on the same page as the review and gets read along with it. This has three effects:
- You correct facts. If a customer writes "the electrician from Weinfelden", but your business is based in Bürglen, your reply sets this straight without touching the review.
- You add the missing service detail. "Glad the wallbox is working well. We install charging stations across the whole Weinfelden district." One sentence that gives the review the service and region it was missing.
- You show how you handle criticism. A factual reply to a two-star review that names the problem and describes the solution is stronger evidence of reliability than ten five-star reviews without any text.
Reply within a few days, with the company name in the signature, without stock phrases. A reply repeated word for word thirty times provides the same information thirty times over, which is to say, none.
Spread across platforms
A model searching for evidence finds more when that evidence sits in several places and agrees. For a Swiss SME, these matter most:
| Platform | Who reviews there | What it contributes for AI |
|---|---|---|
| Google Business Profile | Most customers, because it appears in the search result | Largest volume of text, directly linked to the map and opening hours |
| local.ch | Customers who found the business through the directory | Swiss directory source with address, category and rating in one data record |
| search.ch | Similar to local.ch, often an older audience | A second independent confirmation of name, address, phone number |
| Trade portals (trades, hospitality, health) | Customers with a specific job | Evidence of the service within a relevant topic area |
| Social networks | Few, mostly regular customers | Small amount of text, but another signal that the company exists |
This spread doesn't mean collecting the same number of reviews everywhere. It means your name, address and phone number are written identically everywhere, and that real text appears on the two or three most important platforms. Optimise your Google Business Profile describes how to set up your profile so it works as the main source.
Swiss peculiarity: local directories
In Germany, Google dominates reviews almost entirely. In Switzerland, local.ch and search.ch also matter, holding address data, categories and reviews in a single data record. For queries that name a location, they regularly turn up among Perplexity's sources. Whether other models use them the same way can't be proven, but the fact that they're public, structured and Switzerland-specific suggests they do.
So check two things: is your business listed there under the right category, and does the spelling match your website and Google? Which other listings are worth it is covered in Business directories in Switzerland. How to check in Perplexity which sources are actually being drawn on for your queries is explained in Perplexity for businesses.
What you can do yourself
- Ask specifically. After approval, send a link with the request: «Two sentences about what we did and how it went.»
- Reply within a week, with the service and location in the text, without stock phrases.
- No incentives, no bought reviews. Both break the platform rules and get spotted by models, because bought reviews tend to resemble each other.
- Don't mark up your own reviews. Google doesn't allow review markup for ratings that a business publishes about itself. Structured data for reviews belongs on platforms, not on your service page.
- Quote reviews on the website, with first name, location and service, as plain text. That way the evidence sits exactly where the AI reads about your services.
What reviews can't do
Reviews prove that you exist and what customers know you for. They don't answer technical questions. If someone asks what a heat pump in an older building costs, the best review won't help; what counts is whether your website gives the answer. Both together, the answer on the website and the evidence from reviews, make a business suitable for a recommendation. Whether this already works for you is shown by the free AI check in about a minute.
Frequently asked questions
Does the number of reviews or the average matter more?
For a language model, probably neither on its own, but the text. Twenty reviews that specifically describe what was done provide more evidence than a hundred saying «Great!». The weighting isn't proven, but it can be observed that AI answers quote traits that appear repeatedly in reviews.
Should I have negative reviews removed?
Only if they breach the platform's guidelines, for example because they come from someone who was never a customer. A piece of criticism answered factually is a sign to readers and models that the business is genuine and deals with problems. A profile without a single criticism, on the other hand, tends to look less credible.
Can I give customers something in exchange for a review?
Google's guidelines prohibit incentives for reviews, and other platforms handle it similarly. You can ask, but not reward. Better than a voucher is a specific request at the right moment, for example straight after approval with a link.
Do reviews on local.ch or search.ch actually help?
Both are large Swiss directories with their own reviews and often appear among Perplexity's sources for provider questions with a location. Whether a model draws on them in a specific case can't be guaranteed. But matching details there and on Google reinforce each other.