Where we started
We ran an audit on 99francs.agency in June 2026 and the baseline was poor. Google showed no top positions for our commercial category queries. ChatGPT and Perplexity built every "best design subscription" answer from other providers and never mentioned 99 Francs, because those answers come from listicles and review platforms where we did not yet exist.
What we changed
- Unified the canonical host, so every schema, sitemap and canonical URL agrees on one entity address.
- Rebuilt the Organization schema and added direct-answer blocks to 20+ pages, so each opens with a quotable definition.
- Published llms.txt, and made every page available as Markdown at the same URL for clients that ask for it.
- Published buyer guides and comparison pages around real decision-stage questions, and connected them with internal links.
What we measured
| Metric | Result | Period |
|---|
| Google Search impressions, first 31 days | 59 before, 479 after | Two equal 31-day periods around 7 June 2026 |
|---|
| Queries with impressions | 1 before, 73 after | The same two periods |
|---|
| Google Search impressions, three months | 6,316 | 19 June to 18 September 2026 |
|---|
| Impressions in Google's AI features | 259 | 19 June to 18 September 2026 |
|---|
| AI-assistant sessions vs organic search | About 1.9 times as many | 23 August to 19 September 2026 |
|---|
In the first 31 days one qualified lead told us directly they found 99 Francs through an LLM. Most AI-assistant visits landed on case studies, and in the four weeks to 19 September those visitors engaged at 63%, against 19% for direct traffic.
What is not guaranteed
- Causation. Several changes shipped together, so we cannot say whether schema, internal links, new pages or llms.txt did the work.
- Mentions. Search Console cannot measure how often an LLM names you. The LLM lead is one verified attribution, not a share of voice.
- Special files. Google says sites do not need new machine-readable files or special markup to appear in its AI results. We keep llms.txt and Markdown because other agents use them; the gains came from clearer pages.
- Clean traffic data. The AI Assistant channel in GA4 includes some of our own checks.
What a small company can take from this
In our data the pages that answered one buyer question in one market earned most of the impressions, far more than broad service pages, and comparison pages were among those shown most inside Google's AI answers. Start with clear pages for the questions your buyers ask, and with proof that a machine can read. Our approach to this work is described on AI SEO and AEO; the experiment itself is in Part 4 of the series.