More and more customers no longer click through ten results: they ask ChatGPT or Perplexity a question, read a synthesised answer, and contact one or two of the companies mentioned. For a Swiss SME this moves the playing field. Being on page one is no longer enough; you have to be among the sources the machine considers worth quoting. This guide explains how these engines select their sources, what to change concretely on a website, how to measure your mentions, and what genuinely does not work.
AEO, GEO: what are we talking about
The acronyms keep multiplying, but the idea is simple. Answer engine optimisation means making your business easy to cite for a system that produces a written answer rather than a list of links. It covers three things: being findable at the moment the engine looks for sources, being unambiguously understandable, and being corroborated elsewhere well enough to look trustworthy.
This is not a break with traditional search optimisation: the vast majority of the foundations are identical. A slow, badly structured or invisible site will not suddenly appear in AI answers. The difference lies in formats, factual precision and off-site reputation.
How these engines choose their sources
Engines connected to a live index
Perplexity, ChatGPT web search, Copilot and Google's AI overviews all work on the same general principle: they rewrite the user's question into several queries, pull a handful of pages from a search index, then compose an answer using the most usable passages. Three filters apply in sequence. The engine must be able to reach your page, it must find a passage that directly answers the question, and that passage must stand on its own well enough to be quoted without extra context.
In practice, a page that buries the answer inside a long company narrative will be read and then discarded in favour of a competitor page that states a price, a lead time or a definition in two sentences. Organic ranking influences selection but does not determine it: it is common for a page ranked fifth or eighth to be cited because it is clearer than the top four.
What the model already knows
When the user does not trigger a web search, the model answers from its internal knowledge. There, your own site matters less than what is said about you elsewhere: directories, local press, forums, reviews, partner sites, professional profiles. A business mentioned consistently across a dozen external sources has a far better chance of existing in that memory than one whose information lives only on its own domain.
Structuring pages as answers
The central rule is simple: every page should contain self-contained blocks that answer a precise question in two to four sentences, immediately after a heading that poses the question. No introduction before the answer, no implicit reliance on the previous paragraph.
- Write subheadings as real questions, phrased the way a client would ask them.
- Put the direct answer right underneath, then expand afterwards.
- Give verifiable elements: price ranges in CHF, lead times, areas covered, conditions.
- Repeat the company name and the town in the body text, not only in the footer.
- Date your pages and update the figures instead of leaving stale information.
This way of writing has a useful side effect: it also improves human conversion, because a visitor in a hurry finds the information without hunting for it. We apply this structure to every page produced as part of our content production work.
Structured data
Schema markup guarantees no citation, but it removes ambiguity about who you are. For a Swiss SME, four types cover the essentials: Organization or LocalBusiness for identity, address, opening hours and contact details; Service for each offering; FAQPage for genuine question blocks; and Article for editorial content, with author and update date.
Two points matter more than the rest. First, consistency: the name, address and phone number declared in your markup must be identical everywhere, including on your Google business profile and in directories. Second, honesty: marking up fictitious reviews or services you do not provide backfires, both with classic search engines and with the verification systems answer engines rely on. The rest is standard groundwork, which we handle as part of search optimisation in Geneva.
Off-site presence, the most underestimated factor
On questions such as best agency in Lausanne or reliable plumber in Geneva, answer engines lean heavily on third-party sources: comparison articles, sector directories, regional press coverage, public discussions, review platforms. Your own website almost never decides the outcome on its own, because an engine looks for outside confirmation before recommending a company.
For an SME, the realistic programme comes down to a few durable actions: a complete and up-to-date Google business profile, a clean presence in the serious Swiss directories of your sector, regular and authentic customer reviews, one or two contributions a year to a local media outlet or blog, and partners who mention you by name. It is slow, but it is what builds a reputation a machine can read. The same logic applies if your market sits around Lake Geneva and you are working on visibility in Lausanne.
Measuring your mentions in AI answers
Measurement is still imperfect, but it is not impossible. Four approaches combine well.
- Build a list of fifteen to thirty questions your clients genuinely ask, then test them monthly in ChatGPT, Perplexity, Google and Copilot, noting whether you are cited, how you are described, and which competitors appear.
- Isolate assistant traffic in your analytics: sources such as chat.openai.com, chatgpt.com or perplexity.ai show up as referrers and are easy to track.
- Watch your server logs to confirm that answer engine crawlers actually reach your pages and are not blocked by your robots.txt file or a firewall.
- Systematically ask new prospects how they found you. In an SME, that question asked on the phone often gives more reliable information than any dashboard.
The right indicator is not an abstract score but a trend: moving from two to eight citations across your thirty target questions in six months is a concrete, verifiable result.
What does not work
Several promises circulate and deserve to be set aside. Publishing twenty automatically generated articles a month does not create citations: answer engines converge on specific, verifiable sources, and generic content with no original information is exactly what they discard. Adding an llms.txt file is not enough on its own either, as long as the format is not universally read; it costs nothing, but it does not replace clear pages.
Another dead end is stuffing pages with brand mentions or sentences designed to influence the model. Those attempts are detected and gain nothing. Finally, be wary of services sold as a guarantee of appearing in ChatGPT. Nobody controls the output of a generative model; what you can do is substantially raise the probability of being picked, by making your information accessible, precise and corroborated.
For a Swiss SME the roadmap fits in one sentence: say verifiable things, say them early on the page, and make sure other people say them too. It is less spectacular than a technical trick, but it is what works today and what will still work once the engines have changed again.
