How to win when your search traffic disappears

It started quietly. A dip in the algorithmic curves. Nothing dramatic –just enough to create a nagging sense of unease. Site traffic that had been predictable for years started moving downward. 30% in some segments. 35% in others. So your marketing team does what it’s always done: Reviewed rankings, campaigns, conversion paths. Nothing looks broken. The budget is intact. Yet fewer people come. The explanation is structural. More and more decisions are now formed before your customer even considers visiting your website. The explanation is structural. More and more decisions are now formed before your customer even considers visiting your website.

Search questions are posed to an AI assistant and the answers arrive ready-made – weighed, packaged. No list of links No impartial tour of options between different providers. In that moment, something decisive has already happened: Someone else has defined reality for your brand category.

From gatekeeper to editor

‍
In the search era, marketing was about being there when the customer was looking. You could win through better structure, clearer messaging and smarter distribution. But large language models (LLMs) don't function as gatekeepers to information. They function as editors. They summarise, prioritise and articulate what appears most reasonable based on what they've learned about how the world tends to work. The central question therefore shifts. Instead of wondering how to get more people to click, you need to ask whether your brand exists at all on the model's internal map of the category. If the answer is no, it's not just the click you lose – you're absent from the reasoning.

A common assumption is that AI can distinguish advertising from editorial content. That's not how it works. Models like OpenAI and Anthropic operate on probabilities. They've been trained on vast amounts of text and become skilled at separating language that informs from language that tries to persuade – not through labels, but through patterns. Persuasive language follows predictable structures, and those structures correlate in the training data with lower informational value.

Paradoxically, the most polished language tends to have a predictive tone and risks becoming background noise.

When language is lived, not scripted, that’s precisely what large language models (LLMs) value weigh upwards.

Why perfection makes you invisible

‍
Are you so “professional” you risk becoming background noise?

‍This striving to ”persuasive” in conventional “marketing-speak” sense creates paradox. For years, professionalism has meant clarity, confident tonality and an absence of hesitation. Excellent for human readers. But to a language model, that same perfection can make text indistinguishable from every other text with similar ambition. It becomes background, and background is rarely cited. In the environments that actually shape the models' worldview, you see something different: people reasoning, hesitating, correcting, coming back to report how things turned out. An engineer on a forum describing why she chose one tool over another – with caveats and nuance – carries more weight for the model than a product page containing the same information framed as a promise. The tone is lived, not scripted. That's precisely what models weight upward.

‍

Your brand is no longer defined by you but it matters more than ever.

In an AI-mediated world, it is no longer the brand itself that primarily defines what it is. The picture emerges from how other people describe their experiences of working with you, choosing you or living with the outcome. The model listens less to your claims and more to the collective echo. But – and this is crucial – that doesn't mean brand work loses relevance. Quite the opposite. If your organization is clear about what it stands for, truthful in what it delivers and consistent over time, you give the world a language to use about you. Without that clarity, you can be as good as you like and still be described in generic terms that don't distinguish you from competitors. Your brand doesn't become less important. It changes function – from controlling the message to shaping the language others spontaneously use about you.

“Without brand clarity, you can be as good as you like and still be described in generic terms that don’t distinguish you from comptetitors.”

This means brand strategy and delivery move closer together than most organisations are used to. It’s a harder game – but a fairer one, because the company that actually delivers on its promises holds an advantage no campaign budget can buy.

‍

What sticks?

‍
The time dimension adds complexity. Foundation models change slowly. New perceptions need to find their way into future training runs and it can take months before they settle. Meanwhile, systems that pull current information from the web can react within weeks. New tests, reviews and discussions start appearing in AI responses surprisingly fast. But even there, the same rule applies: isolated efforts rarely stick. What endures is what recurs. One voice is an echo. Many voices become reality.

The PR agency argues for more editorial presence. The SEO expert talks about structure and semantics. The content team wants to produce more, faster. They all hold part of the truth – and they all conveniently recommend more of their own medicine. There is no method that will reliably make a language model start favoring you. The landscape is shifting too fast for guarantees. But it is possible to identify directions that consistently work better.

There is no guaranteed way to make a language model favor your brand. The landscape moves too fast for certainty. But some directions consistently work better than others. Below are some practical actions you can start with.

Make your expertise available beyond your own channels

Your thinking needs to appear in places you don't control:
‍
• Industry forums
• Podcasts
• Technical communities
• Independent publications
‍
Independent mentions carry disproportionate weight. Your own blog or newsletter isn't enough.

What can you do about it right now?

1. Reason instead of proclaim

‍‍
Content that shows how you think – including caveats, trade-offs and limitations – is picked up more often than polished assertions. For brands invested in a perfectly tuned voice, this may feel counterintuitive. But here, a human voice beats a perfect one.

‍

2. Earn the conversation, don't buy it
‍‍
Campaigns can generate text. But models detect the difference between stories written to win something and stories written because they want to be told.

‍

3. Think in layers of time
‍
AI systems operate on different clocks:
• Foundation model training (long-term influence)
• Search-based retrieval (mid-term visibility)
• Real-time fetching (short-term mentions)

Your strategy needs to address all three:
‍
Build presence in material that trains future models
Appear in sources retrieval systems rely on
Generate ongoing, organic mentions

‍

4. Measure what the models actually say about you

‍
Ask AI models the same questions your customers ask
Are you mentioned? In what context? Compared to whom? It's not rocket science – it's what we are systematising right now. New trails are bringing to light insights that traditional SEO analysis misses entirely.

‍

Don’t lose sleep. Let’s solve it together‍
Are you losing sleep over declining traffic? Relax, it not be a sign of waning interest. It may be a sign that the summary is happening earlier – and if you're not in it, no visit is needed. AI rarely recommends whoever shouts loudest. It recommends whoever appears most natural in how reality tends to be described. Future competitiveness won't be decided by the next campaign, but by the work of consistently delivering things worth passing on.

‍