Why deal with AI answers now, and not in a year

The moment someone decides to contact you, apply, buy or invest is preceded by a search. Until recently that search produced a list of links; today, more and more often, it produces an answer written by a model. Whoever has not worked on the sources the model reads is not in that answer, or is there with the wrong version of the facts.

The shift, in numbers

The data below come from public sources from 2025-2026, mostly US or global. They should be read as a trend, but the trend is unambiguous.

• Adoption: ChatGPT has over 900 million weekly users (February 2026); the Gemini app exceeds 750 million monthly users; Google's AI Mode has passed one billion monthly users.
• Search replacement: AI Overviews appear on about 48% of tracked searches; about 60% of searches end without any click; Gartner forecasts a 25% drop in traditional search volume in 2026.
• Depth: a session with an AI assistant lasts 6 minutes on average against a few seconds on Google, and an average question has 23 words against 4. People explain the context and trust the answer.
• B2B: about 90% of B2B buyers use generative AI in the purchasing process, and 17% of the buying journey is spent with suppliers. The rest is independent research, increasingly with AI (Forrester and Gartner, 2025).
• Talent: 21% of candidates use AI to research a company before applying (Insight Global, 2025).
• Traffic: the traffic sent to websites by AI is still just over 1% of the total, but grows by about 1% a month, with ChatGPT generating almost 90% of it.

What it costs not to be there

The right question is not "how much traffic does AI bring", but "how many decisions go through an AI answer without you knowing". Some concrete examples we see in assessments:

• A missed application: a senior profile asks ChatGPT what it is like to work at a company; the answer cites a restructuring from four years ago and dated reviews, and does not mention benefits, training or recent awards. The candidate does not show up, and nobody in the company will ever know why.
• A lost shortlist: a buyer asks Perplexity which suppliers to evaluate for a category. The AI names three, on average three or four per answer. Whoever is not among those names does not receive the request for proposal.
• A due diligence off to a bad start: an analyst asks Gemini what risks a company presents. The answer assembles a closed lawsuit, a critical article and a Wikipedia page frozen three years ago.
• A compromised interview: a journalist prepares questions with Copilot and arrives with a narrative the press office thought was over.

In every case the source of the problem is the same: the AI engines read what was available and repeated, not what was true today.

Why now: the advantage of starting first

Three reasons make timing decisive.

Answers consolidate. Models weigh consistency across sources and repetition over time. A company that appears in answers today with the right facts tends to stay there; whoever arrives later has to move a perception already formed, and moving is more expensive than building.

There are few slots. An answer cites three or four names on average. In every category the slots are being assigned now, based on the sources that exist now.

Third-party sources take time. A Wikidata entry, a listing in an industry directory, an interview, a run of recent reviews: these interventions have their own timelines, and models absorb them with weeks or months of delay. Starting today means having an effect next quarter, not tomorrow.

What to do this week

1. Check what the AI engines answer about you: use our DIY test with ten questions, or request the free 48-hour audit.
2. Check the primary sources: Wikipedia if it exists, LinkedIn, the Google listing, Bing Places, reviews. They are the first things models read.
3. Decide the scope: which audience matters most today, in which country. A GEO Assessment starts from there.