A year ago EMARKETER analyst Nate Elliott published a report with a blunt title: AI Will Dominate Search—Just Not in 2026 (October 2025). We are now ten weeks from the end of 2026, so the forecast can be checked against this month's traffic data. It is holding up. For creators and solo builders deciding where to spend their limited hours, that gives a fairly clear split, set out below with numbers you can run on your own site.

The core finding is about time. Using August 2025 Comscore data, EMARKETER estimated that generative AI engines account for just 3.3% of US online discovery time (pp. 3–6). Traditional search engines took nearly 97% (p. 8). That gap existed despite enormous AI adoption: OpenAI reported more than 800 million weekly ChatGPT users, and 60% of US adults had used generative engines to search for information (p. 5).

The report's explanation is that AI use was being added on top of search. Comscore data showed US users visiting generative engines about three times as often as a year earlier and spending nearly four times as long, while Google's search visits and time spent stayed flat (p. 6). In raw terms, ChatGPT, Claude and Perplexity together received 481.1 million US web visits in August 2025, against 29.20 billion for Google Search (p. 6). SparkToro data cited in the report found people 2.5 times more likely to use a search engine monthly than a generative engine (p. 6).

The forecast itself has two parts. First, AI discovery merges into search: Elliott expects that by 2027 "we won't be able to tell most generative engines and search engines apart" (p. 4), with Google defaulting longer, complex queries into AI Mode within a year (p. 7). Second, the commercial gap: OpenAI said only 2.1% of ChatGPT conversations mention purchasable products, against 14.5% of Google searches that SparkToro classes as commercial (p. 7). The report's advice to brands is to keep SEO funded and start measuring AI visibility now.

Two sets of data reported this week line up with both parts of the forecast.

On volume, Marie Haynes analysed 12 months of GA4 data from 100 websites, more than 145 million sessions. Search Engine Land reports that ChatGPT drove just 0.62% as many sessions as Google organic search, or 0.73% with all LLMs included. ChatGPT visitors did convert better, 13.8% against 3.02% on the 55 sites tracking conversions, but Google organic still drove 27 times as many conversions as all LLMs combined.

On the merger, Semrush tracked 400 US commercial keywords from September 2025 to August 2026. PPC Land reports that the share showing an AI Overview rose from 45.75% to 88%, and for "reviews" queries from 30.14% to 89.04%. That is US desktop only, and it records only whether an Overview appears, with no click data. The report had already flagged the click cost: Pew found people about half as likely to click out of Google when an AI Overview appears (p. 7).

My take: the AI search that matters to your income in 2026 is the one sitting inside Google. ChatGPT is a small, high-intent side channel. Put most of your effort into pages that rank and get cited in Google's AI answers, and a smaller, measured slice into standalone AI engines.

A sensible default for a creator business with existing search traffic is roughly 80% of discovery effort on search (including AI Overviews and AI Mode) and 20% on standalone answer engines. Adjust it once you have your own numbers from the test below. The split matters less than what you do in each bucket.

Search bucket. Keep publishing pages that answer specific, longer questions. Queries are already getting longer: 69% of Google searches in June 2025 had three or more words, up from 58% in January (p. 7). Write for the full question your buyer types, lead with the direct answer, then add the detail an AI summary can't hold: your prices, your templates, your test results. Those details are what earn the click after the Overview.

Answer-engine bucket. Treat ChatGPT, Perplexity and Claude as a referral source to measure and nurture. We covered how to become a cited primary source in publishing your own data on living URLs, and the formatting side in the LLM citation engine, so I won't repeat either here. The point today is sizing: give this bucket effort in proportion to what it returns.

Illustrative maths, not results. Use your own analytics in place of these assumptions.

Example 1: a $49 template pack. You get 10,000 Google organic sessions a month and 1% buy, so search earns $4,900. Apply the Haynes ratio of 0.73% and you'd expect about 73 AI-referred sessions. Even if those convert four times better at 4%, that is about three sales, or $147. Doubling your AI traffic adds roughly $147 a month; a 10% lift in search traffic adds $490. Spend accordingly.

Example 2: an affiliate review site. If 89% of your "reviews" queries now trigger an AI Overview and click-through roughly halves on those pages, a page earning $1,000 a month could fall towards $500 without any ranking change. The fix is to give people a reason to click: original test photos, a downloadable comparison sheet behind an email field, a price tracker. Each one turns a summarised visit into a subscriber you own.

Example 3: a consultant or service seller. Low volume, high value. If five AI-referred visits a month include one $2,000 client, the 13.8% conversion figure matters more than the traffic share. Here the answer-engine bucket can deserve more than 20%, as long as your tracking shows the leads.

The report's first recommendation is to create a baseline of existing AI visibility (p. 7). Here is a solo version that takes about two hours.

Repeat the 20 queries monthly in the same spreadsheet. If Google does start defaulting longer queries into AI Mode, as the report expects, you'll see it in your own data first.

A caveat on all of this: the EMARKETER figures are mostly US and from 2025, and the 2026 studies use small or self-selected samples. They point the same way, which is why I'd act on them, but your analytics should have the final say.

From today's Daily Briefing →