Find the AI Assistant Sending You Signups in One Afternoon
A one-afternoon referrer audit tells you which AI assistant is already sending signups, so you can spend the next week on the one that works.
The useful question is which assistant already sends signups
You open GA4 and see a handful of signups from AI assistants. Treat AI discovery as a leaky pipe you can inspect, not a growth channel to chase everywhere. You do not need a full AI strategy. You need the assistant that is already putting your product in front of people.
Previsible's study sampled 166 GA4 properties from November 2024 through May 2026. Its dataset held 6.77 million AI-assistant sessions, and ChatGPT accounted for 92.4% of them. Monthly AI referral sessions in that dataset grew 9.9 times from November 2024 to May 2026.
In December 2025, ChatGPT made up about 84% of the measured AI referral traffic, ahead of Perplexity, Gemini, Copilot, and Claude. That spread tells you where most AI traffic sits. It does not tell you where your signups are.
If you have no ads budget, paid placement is not the first move. OpenAI made a self-serve ChatGPT Ads Manager available on July 22, 2026, and sponsored placements reached an estimated 51% of US replies in seven markets. Your early users are more likely to come from one assistant that already recommends you than from a paid test you cannot afford to run.
One afternoon tells you which assistant is working
Run a referrer audit before you build a campaign. Do the three steps in order, because the last one depends on the first two.
- Filter GA4 signups by AI referrer and write down the assistant with the highest count.
- Open the landing pages for those AI referrals and note whether they are product pages, docs, or internal search results.
- Pick the assistant with the most signups and make it the one you will optimize citations for next week.
Aggregate AI traffic can mislead you. The assistant that dominates the market may not be the one sending you signups. You are optimizing for your own funnel, not the category.
A referral to your homepage is weaker than a referral to the page that answers the question. If the assistant sends people to an internal search page, that is a signal about what they were asking. It is also a clue about the page you should make easier to cite.
Your next week is too short for a multi-assistant campaign. You need a target, a set of questions, and a set of pages to make citable.
The next week belongs to the assistant that already works
Once you have the name, do the unscalable thing. Read the pages it sends people to. Rewrite the first paragraph so it answers the likely question directly. Add the missing detail a user would need to decide.
Then check whether your product page, docs, or comparison page is the page that should be cited. If it is not, make it the page. A short, direct answer is more useful than a clever narrative.
Do not spread the week across every assistant. Perplexity's monthly referral sessions hit 17,507 in March 2025, then dropped to 6,788 by May 2026, a 61% decline. Claude first overtook Perplexity in monthly referral sessions in March 2026 and kept the lead afterward. The ranking is moving, so do not build for all of them.
Use the assistant itself to test the question. Ask it the question your user would ask, and see whether your product appears. If it does not, change the page until it does. It is the only AI referral task that pays off before you have scale. Finish that before you write another launch post.
The page needs to be worth citing
For an early-stage product, the working channel may be a single assistant and a single page. Make that page answer the question in the first few lines, not after a brand story. Name the use case, the limitation, and the next step. Keep it easy for an assistant to summarize without losing the point.
If the page is a comparison page, make the comparison concrete. A docs page needs a clear setup path. A product page needs an obvious outcome. The assistant is answering a question, and your page either helps it do that or it does not.
Then do the unscalable follow-up. Find the people who asked the question in public forums, communities, or support threads, and send them the page with the answer. If the assistant is already recommending you, your job is to make the recommendation easier to trust. Send the page, then check whether the next signup came from that assistant.
