AI assistants are sending traffic — and almost nobody is optimising for it yet.

Search engines dominated web discovery for two decades. Social media became the second channel. A third channel is now forming: AI assistants answering questions and recommending specific products, brands, tools, and content to millions of people every day.

ChatGPT, Claude, Gemini, Perplexity, and their derivatives are handling queries that used to go to Google. And unlike Google, these systems make specific recommendations. They name products. They suggest brands. They answer “what should I use for X” with a direct answer — not ten blue links.

The signal: early-adopter brands and products are already appearing in AI-generated recommendations. Traffic from AI referrals is appearing in analytics. The operators who recognise this now have a 12–18 month window before the mainstream catches up.

Observable evidence from the field, right now.

This is not a prediction. These are things happening today that most operators have not yet connected into a coherent picture.

• AI referral traffic is appearing in Google Analytics. Site owners are seeing sessions attributed to chatgpt.com, perplexity.ai, and similar sources. Volume is small now — but it is growing month-on-month and the trend is consistent.
• AI assistants are citing specific brands by name. Ask any major AI about tools, services, or products in a category and it returns named recommendations. Those names are not neutral — they reflect training data, content quality, and how well-represented a brand is in the sources the model was trained on.
• Perplexity is building a shopping layer. Perplexity has announced and is rolling out product recommendations with direct purchase intent, backed by merchant partnerships. This is the first explicit monetisation of AI recommendation traffic.
• The query types are different. AI queries skew toward decision-support: “what should I use”, “what’s the best option for”, “compare X and Y”. These are the highest-intent queries in existence — the ones that used to be worth the most in paid search.
• Traditional SEO signals are not the same signals here. Backlinks, domain authority, and keyword density — the mechanics of Google SEO — do not directly map to AI recommendation influence. Something different is being optimised for, and the playbook is not yet written.

This is a distribution shift — and distribution shifts create the biggest windows.

Every major distribution shift in the internet era has created a window where early movers built durable advantages. Early Facebook pages built audiences that cost nothing. Early YouTube channels captured attention before the platform was crowded. Early newsletter writers built lists that are now worth millions. The window was always short, and the operators who moved early always won disproportionately.

AI recommendation traffic is the next shift. The platforms are built. The user behaviour is forming. The queries are happening. What does not yet exist is a coherent strategy for how to show up in AI recommendations — because most of the people who would build that strategy are still focused on Google.

The operators who move now are optimising for a channel that is growing fast, has almost no competition yet, and handles the highest-intent queries on the internet. The operators who wait will be paying to compete in a channel that everyone has figured out.

Five specific opportunities that are open right now.

The opportunity is not one thing. It is a cluster of openings, each at a different stage of accessibility and competition.

1. Become the named brand in your category. AI models learn from the internet’s existing content. The brands that are most clearly, consistently, and authoritatively described as the best option for a specific use case are the ones getting recommended. Define your category sharply, own it in writing, and make it impossible for an AI to talk about your category without mentioning your brand.
2. Build content that answers the exact questions AI assistants field. “What is the best X for Y” queries are your target. Create genuinely useful, specific, well-structured answers. Not SEO filler — actual operator intelligence that AI models can cite, quote, and recommend from.
3. Get cited by authoritative sources AI models trust. The training data for these models skews toward authoritative, well-referenced content. Getting your brand, product, or methodology cited in publications, directories, and content hubs that AI models treat as authoritative is a leverage point that most operators have not started working.
4. Build an AI-native presence on Perplexity and similar platforms. Perplexity is building explicit merchant and publisher relationships. Being early to those programmes, before they are oversubscribed, is a straightforward tactical move.
5. Track and measure AI referral traffic now. Set up the measurement infrastructure before you need it. UTM tagging, referral source tracking, and a simple dashboard for AI-attributed sessions will give you real data on what is working — and a head start on the operators who are still ignoring this channel entirely.

This shift does not reward everyone equally.

A specific action sequence for the next 90 days.

This is not a strategy document. This is what I would actually do, in order, starting tomorrow.

1. Instrument AI referral traffic today. Log into Google Analytics or your analytics tool of choice. Set up a custom segment for sessions from chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, and bing.com (Copilot). Set up a weekly report. You need the data before you can optimise anything.
2. Define your category claim in one sentence. Not your tagline. A specific claim: “The best [product type] for [specific user in specific situation].” This is the sentence you want AI models to reproduce when someone asks about your category. It needs to be in your site copy, your about page, your product descriptions, and every piece of content you publish.
3. Publish five genuinely useful “what is the best X for Y” articles. Not keyword-stuffed. Actually useful. The kind of answer a knowledgeable friend would give. These are the documents that AI models will draw on when answering recommendation queries. Make them citable: specific, well-structured, with concrete recommendations.
4. Pursue three authoritative citations. Identify the directories, roundups, and publications in your category that AI models trust. Get listed, cited, or featured. One authoritative citation in the right place is worth more than a hundred low-quality mentions.
5. Test your AI visibility now. Open ChatGPT, Claude, and Perplexity. Ask them the questions your ideal customers ask. Do you appear? If not, what appears instead? That gap is your brief. Work backwards from the recommendation you want to receive.

Honest risk assessment. Every signal has a failure mode.

A conviction score of 9 is not certainty. Here are the specific conditions that would reduce or eliminate this opportunity:

None of these scenarios are likely in the 6–18 month window this signal is active. The recommendation remains: act now, build for the open window, review in 12 months.

The upstream signals that will shape how this plays out.

These are the indicators we’re monitoring that will determine the size and duration of this opportunity window.

When any of these signals move materially, it will affect the time horizon and conviction score for this opportunity. We will update this report if the signal changes significantly.