Last April we placed a so-called llms.txt file on the first of our client websites — technically explaining what the page is about, to help large language models incorporate it. At the time it was an experiment we didn't even know how to measure.
Twelve months later, the situation has changed. For 12 of our 28 e-commerce clients, 18–34% of organic traffic now comes from sources that are not the classic Google click. A person asks ChatGPT about "the best ergonomic chairs in Latvia" — ChatGPT responds with specific product names and a link to our client's website. The click arrives with high intent already attached.
This is moving fast. And we still don't know if it will move the way we think — but we know enough to say: the classic SEO approach no longer answers the fundamental question.
We no longer optimise a page to appear in Google. We optimise knowledge to appear in the answer — regardless of which machine reads it.
— — A. Bērziņš, MH Journal № 42
What we actually measured.
For 12 months we tracked daily how often ChatGPT, Perplexity, Gemini, and Copilot mentioned our clients' websites in their answers. Technically this happened in three ways:
(1) Synthetic prompts — we sent each LLM 80 questions per month about our clients' products and industries and recorded whether and how the client's name appeared. (2) Server logs — we analysed which bots had visited from which OpenAI and Anthropic IPs. (3) GA4 attribution — referrer traffic from chatgpt.com, perplexity.ai, gemini.google.com.
Here is the simplified summary of what we saw by quarter:
| Quarter | LLM citations (Σ) | LLM referrer traffic | Classic organic | LLM share |
|---|---|---|---|---|
| 2025 / Q2 | 214 | 1 480 | 18 600 | 7.4% |
| 2025 / Q3 | 386 | 3 120 | 19 100 | 14.0% |
| 2025 / Q4 | 512 | 4 980 | 21 400 | 18.9% |
| 2026 / Q1 | 748 | 7 320 | 22 900 | 24.2% |
| 2026 / Q2 (through May) | 892 | 9 480 | 23 500 | 28.7% |
The number we want to underline: this is not a classic decline in organic traffic. Classic Google traffic has grown slightly. A new layer has simply been added on top — and in that new layer our clicks appear with less effort but higher intent.
Why the old SEO approach no longer works fully.
Classic SEO works like this: your content is indexed, ranked by relevance and authority, and Google shows your website as a clickable blue link. In large language models, everything works differently.
An LLM doesn't index your page — it embeds your facts into its knowledge. The difference is fundamental: to appear in an answer, you need (a) a citable position (ideally as a factual authority in your industry), (b) structured content (so the LLM understands what to extract), and (c) frequent mention in other citable sources (because LLMs work with language associations, not PageRank).
You no longer optimise a website — you optimise how the rest of the internet talks about your company.
This means that digital PR, guest publications, and publishing genuinely valuable data — which five years ago seemed "nice to have" — are now a technical requirement. If other places don't talk about you, LLMs will never know you exist.
↑ this is the same principle discussed in our Q1 report
Six practical changes.
This is not a theoretical article. Here is what we are actually changing on client websites, based on the data from the past 12 months:
1. llms.txt file and structured data. If an LLM doesn't understand what your page is about, it won't try to use it. Schema.org markup is no longer optional.
2. Author bios and E-E-A-T signals. LLMs want to see a human behind the text. Pseudonyms and "Admin" authorship — that no longer works.
3. Content with proprietary data. You can't compete with GPT on a "what is SEO" article — but you can win with "LLM citation study across 12 Latvian e-commerce websites". Originality is the only currency.
4. Inline citability. Numbers, definitions, frameworks — in a format that allows an LLM to find, extract, and cite them. Tables, lists, and code snippets work better than long paragraphs.
5. Digital PR. Appearing once a month in a major outlet is now a technical SEO action, not just brand awareness.
6. Measurement. Add an LLM referrer segment to your GA4 dashboard. The Looker Studio template we use is available to our clients.
What we still don't know.
Honestly: three things we don't yet fully understand.
Will the LLM citation share continue to grow at this pace? We assume yes, but the baseline in our sample may be specific (e-commerce, B2C, LV/EN markets). Does it apply to B2B SaaS in Scandinavia? We don't know.
Will Google SGE (AI Overviews) replicate the same mechanism? Data on this is currently thin. We can track ChatGPT clicks, but Google AI Overviews often simply don't send the user to the website, because the answer stays inside Google.
How long do classic SEO techniques still work — backlinks, anchor texts, position tracking? We believe that in 2026–2027 they remain useful. But 2030? We wouldn't dare to claim.
Takeaway.
If you are currently thinking about your website's SEO for 2026 and beyond — start with two questions. (1) What are you a citable source for in your industry? If there is no answer — start there. (2) Why should an LLM know about you? If the answer is "because I bought ads" — that is not a reason. That is a signal.
We continue to experiment. The next article on this topic — covering Gemini and Copilot citation patterns — will be published in July.