GEO for a Swiss SME: the first five fixes

Why most Swiss SMEs are invisible to every AI answer

Ask ChatGPT or Perplexity an open question about a trust company, a real estate agency or a private clinic in your city, without naming it. In most cases the business does not show up anywhere in the answer, even when its website has ranked comfortably on Google for years. That gap is rarely a service quality problem. It is a citability problem: nothing in the site's structure gives a generative engine a clean passage to extract and reuse.

A GEO fix is a concrete and verifiable change to a site or its off-site presence that raises the odds a generative engine will pick it as a source, without manipulation and without a guaranteed outcome attached to it. It is not a hidden keyword or a paragraph padded for an algorithm. It is an answer that is simply easier to trust and to lift than the competitor currently occupying that spot.

We already published an overview of GEO and the six levers of visibility in AI answer engines when this blog launched. This piece starts from the opposite end: what does a Swiss SME with no dedicated marketing team actually do first, this week.

Diagnose before you fix anything

Fixing without diagnosing is guessing. Before changing anything, collect fifteen questions your customers genuinely ask, in their own words, and put them to ChatGPT, Perplexity and Google search with AI features turned on. For each one, record three things: is your company cited, which source occupies that spot instead, and is that source more recent or better structured than your own page. The full protocol, with a reusable tracking table, lives in our article on measuring AI search visibility; we will not repeat it here, but none of the five fixes below make sense without this starting baseline.

For a site under fifty pages, this audit takes two to three hours and needs no paid tooling. As a data point: our own domain, launched in early September 2026, still has not accumulated a single query above fifty impressions in Search Console after two weeks. Patience is part of the protocol, not just execution.

Log the results in a simple table, one row per question, one column per engine tested. That document becomes the reference for everything that follows: it is what lets you say, four to six weeks later, whether a fix actually worked instead of relying on a hunch. Many owners skip this step because it feels less productive than acting on the site directly, yet it is exactly what prevents months of fixing things at random.

The first five fixes, in order of impact

These are the five fixes we apply first with a new client, ranked by the ratio between effort and observable effect. Order matters here: adding Schema.org markup before you have natural-language answers worth marking up produces nothing measurable.

1. Answer in plain language within the first two sentences

A generative engine splits a page into self-contained passages before deciding which one to cite. If it has to wade through a paragraph of marketing framing before reaching the fact, it moves on to a more direct source. Take your ten most common customer questions, give each one an H2 or H3 phrased the way a customer would actually ask it, and make sure the first two sentences underneath contain the complete answer, with a figure where possible, without pointing further down the page.

2. Clear the technical path for AI crawlers

Content injected by JavaScript after load, a robots.txt file that blocks GPTBot by mistake, or no llms.txt on a site of several dozen pages: these are three common failures invisible from a regular browser tab. We covered the actual trade-off behind blocking AI crawlers and the client-side rendering trap in two separate articles. For an SME, the practical habit is checking once a quarter that what a crawler receives matches what a human visitor sees on screen.

3. Add the Schema.org markup that actually matters

Not every Schema.org property carries equal weight for an answer engine. For a service SME, three types cover most cases: Organization with complete contact details, FAQPage on pages that explicitly answer questions, and LocalBusiness, or its sector-specific subtype, when the business serves a defined geographic area. We break down which types genuinely matter, and which add nothing, in our article on Schema.org types for answer engines.

4. Build consistent mentions off your own site

A generative engine cross-checks several sources before citing a business. If your name, address and line of business are not written identically across your own site, your business listing, Swiss professional directories and any local press mentioning you, every inconsistency slows down the trust the AI assigns to the whole file. This work is not technical at all: it is a verification and standardisation exercise, directory by directory, that most SMEs can finish in an afternoon or two.

5. Measure again before you touch anything else

The last fix is repeating the initial diagnostic, on the same fifteen questions, four to six weeks after applying the first four. Without that check, there is no way to know which of the four actions actually moved anything, with the risk of rolling out a useless fix site-wide, or dropping an effective one because it was judged too early.

The mistakes that cancel out these five fixes

We keep seeing the same mistakes erase the benefit of an otherwise sound GEO effort:

• Duplicating French content into German or English word for word, instead of adapting it to how each language actually phrases the question: a generative engine treats each version as a separate signal, and a mechanical translation sends a weak one.
• Inflating track-record figures to look more established: an engine that cross-references an unverifiable claim against a contradicting source lowers trust in the entire site, not just the offending page.
• Copy-pasting generic Schema.org blocks from another industry, which creates a mismatch between markup and visible content, a stronger negative signal than having no markup at all.
• Treating the work as finished after the first round of fixes, without ever coming back to measure, which makes it impossible to tell a real improvement from ordinary model fluctuation.
• Treating the five fixes as a one-off project instead of a quarterly routine: a site that changes structure, adds pages or switches technical providers can silently reopen a block that was already fixed, with nobody noticing.

When these five fixes are not the priority

These fixes assume a minimal baseline: a site that Google actually indexes, no major technical block, and a market where prospective customers genuinely ask informational questions before buying. Three situations make this work premature. If your site is not indexed at all or suffers from blocking technical errors, fix conventional indexing first: no answer engine cites a page it cannot find. If your business sits in an ultra-niche market where informational search volume is close to zero, say an industrial subcontractor with five named accounts, time spent on GEO will return less than direct outreach. And if your real bottleneck is conversion rather than visibility, meaning you already get inquiries but struggle to turn them into signed work, these five fixes will not touch that problem: they act upstream of demand, not on your closing rate. In that case, budget is better spent reworking the inquiry-to-contact path or sales follow-up than restructuring pages.

How long before you see a measurable effect

There is no guaranteed timeline, and any precise number attached to this question deserves skepticism. A 2026 guide published by HubSpot on generative engine optimization places the first observable effects between four and eight weeks after the fundamentals are in place: a complete business listing, FAQ markup, plain-language answers near the top of the page. That matches what we see in practice: technical fixes, such as crawler access, can show an effect within one to three weeks when a real block existed, while building consistent off-site mentions takes longer because it depends on third parties, directories or press, that republish on their own schedule.

What to take away

A GEO fix worth making for a Swiss SME is a verifiable change that makes an answer easier to extract and cross-check for a generative engine, never a trick aimed at an algorithm. Diagnose before acting, fix in order of impact rather than in parallel, and re-measure after four to six weeks: that sequence, more than any single trick, is what separates the Swiss SMEs starting to show up in AI answers from the ones still waiting.