All articles
    Strategy
    April 18, 2026

    Why your next customer won't search — their AI will

    Generative AI is replacing the search bar. Here's how to make sure your brand is the one being recommended.

    The default mental model of "a customer types into Google, clicks a blue link, lands on your site" is dying faster than most marketing teams realise. In the last 18 months, the front door of the internet has quietly moved. ChatGPT crossed 800M weekly users. Gemini is built into every new Android device and Workspace tab. Perplexity is replacing Google for an entire generation of knowledge workers. Copilot ships inside Edge, Windows and Office.

    For your next customer, the question is no longer "which website should I visit?" It is "which product should I buy?" — and an AI is answering it for them, often without ever rendering your site.

    What changes when AI becomes the search bar

    Three shifts compound at the same time:

    1. Zero-click answers become the norm. The model summarises, compares, and recommends — the user gets what they need before any click happens. Even when a citation is shown, click-through rates collapse to 1–3%.
    2. The shortlist is generated, not crawled. When a shopper asks "what's the best merino base layer under €120 for ski touring?", the model returns 3–5 brands. If you're not in that shortlist, you don't exist for that query — regardless of how well you rank on Google.
    3. Recommendations are reasoned, not ranked. The model isn't matching keywords. It's reading your product copy, your reviews, your schema, your shipping policy, and weighing them against competitors. Weak structured data = weak reasoning = no recommendation.

    Why "good SEO" is not enough anymore

    Classic SEO optimises for a crawler that wants to index a page. Generative engines optimise for a model that wants to understand and quote a page. The two overlap, but the gaps are widening:

    • Schema isn't optional. A page without Product, Offer, Review and Organization JSON-LD is a page the model has to guess about. Models don't like guessing — they cite the page that already gave them a clean answer.
    • llms.txt and robots access matter. If GPTBot, ClaudeBot, PerplexityBot or Google-Extended can't reach you, you're invisible to the assistant pipeline that trains and grounds these models.
    • Conversational copy beats keyword copy. Models reward pages that answer questions in natural language ("How long does the merino crew last?") over pages that stack keywords for a 2014 Google algorithm.
    • Fresh, verifiable facts win. Price, stock, sizes, materials — if these aren't machine-readable and recently updated, the model defaults to a competitor it can verify.

    The new funnel: discovery → grounding → transaction

    The shopper journey is collapsing into three AI-mediated steps:

    • Discovery — the assistant decides whether you make the shortlist for a prompt.
    • Grounding — the assistant pulls from your product page, reviews and policies to justify the recommendation.
    • Transaction — the assistant either hands the shopper to your storefront, or (in the agentic commerce era now arriving) completes the checkout itself via MCP / AP2 / UCP endpoints.

    If you're invisible at step 1, nothing else matters. If you're present at step 1 but unparseable at step 2, the assistant downgrades you. If you're missing the agent endpoints at step 3, you'll watch competitors capture agent-driven GMV while you wait for the click that never comes.

    What to do this quarter

    You don't need to overhaul everything. You need to be measurable, then move:

    1. Baseline your AI visibility across ChatGPT, Gemini, Claude, Perplexity and Copilot for the 50 prompts that actually drive your category.
    2. Fix the structured data on your top 20 revenue pages first. Product, Offer, FAQ and Review schema generate the biggest lift per hour of work.
    3. Open the doors to AI crawlers. Audit robots.txt, ship an llms.txt, and make sure your sitemap is current.
    4. Track share-of-voice per model, not just per keyword. The model that converts best for you today might not be the model that wins next quarter.
    5. Plan for agent endpoints. MCP, AP2 and the emerging UCP standard will be table-stakes within 12 months for any merchant selling something an agent can buy.

    The brands that win the next decade of commerce won't be the ones with the prettiest homepage. They'll be the ones the AI trusts enough to recommend by name — every time, in every model, in every language.

    Your next customer won't search. Their AI will. Make sure it picks you.