• AI visibility

Guide · AI search

What should our AI search strategy be for the next five years?

Plan for a channel that is growing fast but whose rules are still being written. Put most of your effort into what has held across engines and years so far: credible third-party coverage, accurate and findable information, and honest measurement. Keep options open on what is unsettled, such as ads in AI answers, Google’s AI Mode and which assistants win. No study forecasts five years ahead; the research shows a direction and a set of open questions.

By Underneath Team Guide Updated

The short version

  1. Google answered 67% of the same US searches with an AI Overview in 2025, up from 42% in 2024 (Aral, Li and Zuo (opens in a new tab)).
  2. AI assistants already pick products: ChatGPT stated a first-person product preference in 79% of product-recommending answers in a 2026 audit (Uberti-Bona Marin and colleagues (opens in a new tab)).
  3. The channel is fragmented: for the same question, ChatGPT and Gemini shared only 5.4% of the websites they displayed.
  4. In a US field experiment with 1,100 people, making AI Mode the only Google experience cut clicks to outside websites by 18.8 percentage points (Wang and colleagues (opens in a new tab)).
  5. A 2026 review of 45 studies found no optimization technique with a proven, lasting effect across engines on discovery or buyer behavior (Martinez (opens in a new tab)).

Can anyone predict what AI search will look like in five years?

No: the best evidence covers days and months, not years, and the platforms change faster than studies run. Plan for uncertainty, not a forecast.

The main randomized experiment on AI in Google search, by Wang and colleagues (opens in a new tab), measured seven days of use. A Swiss study by Schulte and colleagues (opens in a new tab) tracked German-language prompts on four AI engines for about 45 days. Roughly 65% of cited sources changed from one day to the next. A critical review (opens in a new tab) of the field rates one claim with high confidence: commercial AI engines differ from one another and change over time.

Platform decisions can also flip overnight. Aral, Li and Zuo (opens in a new tab) ran the same Google searches in 2024 and 2025. AI answers on Covid questions went from 1% of answers to 66% globally, which the authors attribute to a policy change. A five-year plan has to survive changes like that.

How fast is AI search actually spreading?

Fast, on every measure we found, though some figures come from industry sources cited secondhand. The trend is clear even where the exact size is not.

In the Aral study, 67% of identical US searches showed an AI answer in 2025, against 42% in 2024. Shopping searches started from a low base but moved fastest: in the seven early countries, AI answers on shopping searches rose 222% in a year. Still, only 13% of shopping searches worldwide showed one in 2025, so commercial searches remain less covered than general knowledge.

Usage is rising outside Google too. The St. Gallen researchers cite ChatGPT growth to about 780 million weekly users by September 2025, from about 100 million in January 2024. They also cite clickstream research in which traditional search queries fell by more than 20% after people adopted AI search. A 2026 position paper (opens in a new tab) cites an AP-NORC poll of 1,437 US adults, in which 60% said they use AI to find information at least some of the time. We could not check these secondhand figures against their original sources.

Is generative search a new distribution channel for companies?

Yes, increasingly: AI assistants now shortlist and recommend products, not just links. But it is a fragmented channel you cannot yet buy into reliably.

In the audit by Uberti-Bona Marin and colleagues (opens in a new tab), ChatGPT stated a first-person product preference in 79% of answers that recommended a product. The same paper cites the largest published measurement of ChatGPT use, in which product and service recommendations were roughly 2% of conversations in 2025. The position paper cites a Salesforce survey in which 39% of 8,350 shoppers across 21 countries used AI for product discovery and related tasks.

Unlike search, the channel has no single front door. ChatGPT and Gemini shared only 5.4% of displayed websites for the same question, with no website in common in 76.7% of comparisons. In our four-assistant study, 66.3% of the options recommended for a question came from one assistant only. For where assistants sit in the buying journey, see our guide to ChatGPT and Google in the customer journey.

What happens to website traffic as AI answers spread?

It falls in the best evidence so far, and Google’s design choices drive the size more than yours. Plan for fewer, later visits.

In the Wang experiment with 1,100 US participants, forcing every search into AI Mode cut clicks to outside websites by 18.8 percentage points. Outside Google, Iannelli and Ai (opens in a new tab) studied a browsing panel; both work for an AI visibility company. They found that 34.1% of sessions involving an AI assistant had no visit to any outside website. For the same people, 19.5% of search-led sessions had none.

So the value of an AI answer increasingly sits in the answer itself: whether you are named, and how. For an estimate of what a default AI Mode would mean for your own traffic, see our AI Mode traffic analysis.

Which parts of the future are genuinely uncertain?

Three are wide open: how AI answers get paid for, whether users accept AI Mode, and which assistants lead. Each could change your plan.

Ads. OpenAI expanded advertising in ChatGPT to 31 European markets in August, according to the product-recommendation audit. A Carnegie Mellon model by Zhang and colleagues (opens in a new tab) suggests that competition between AI engines pushes them toward fewer ads. In their simulations, ad-heavy policies earned more at first but shrank the user base. That is a model, not observed behavior. Our guide on ads in AI assistants covers what is known.

User acceptance. Google presents AI Mode as the future of search, yet in the Wang study it was used for only 0.6% of searches before the experiment. Forcing it on people raised the share who searched on a rival engine by 11.2 percentage points and lowered their trust in Google’s information.

Which assistants lead. With so little overlap between assistants, a plan built around one of them is a bet on that one surviving and staying the same.

Which bets hold up whatever happens?

Credible third-party coverage, accurate information that search systems can find, and measurement you can trust. Each is useful under every scenario above.

Earned media is the most consistent pattern across engines. In a 2025 study by Chen and colleagues (opens in a new tab), ChatGPT drew 93.5% of its cited sources for well-known brands from earned media, such as reviews and publishers, and Claude 87.3%. The critical review rates two levers with high confidence: how closely content matches the question, and where it sits among the material the AI reads.

Tactics are the weak bet. The same review of 45 studies found that no technique shows a stable, long-run effect across engines on discovery or buyer behavior. It rates the link from AI citations to clicks, conversions or revenue as very low confidence. A framework by Kato and colleagues (opens in a new tab) starts from the same gap: standard marketing data do not record how often buyers see and notice a firm’s name in AI answers.

What should you do about it?

Treat AI answers as a lasting channel with an owner and a yearly review. Build durable assets rather than chasing tactics. In practice:

  1. Give the channel an owner. Someone should track AI visibility, ads and platform changes across engines and markets.
  2. Invest in durable assets. Earn coverage in the publications and review sites AI engines cite, keep facts about your company accurate everywhere, and keep pages easy for search systems to find.
  3. Measure as a range, not a rank. Track several assistants and countries, repeat questions over time, and report how often you appear.
  4. Build business measurement now. Use comparison groups and stated start dates so you can tell your own effect from the market’s growth.
  5. Set triggers, not forecasts. Decide in advance what you will change if Google makes AI Mode its default, if assistants open self-serve ads, or if one assistant clearly leads your category.
  6. Keep search fundamentals. Search still opens many buying journeys, and AI engines run searches of their own.

If you want help building a multi-year plan, see our generative engine optimization service.

What does the research not tell us yet?

It cannot say where AI search will be in five years, or which strategies pay off over that span. The main gaps:

  • No long-run study exists: the causal evidence covers a week, and most audits cover days to months.
  • Several growth figures, including weekly users, survey adoption and declines in traditional searching, come from industry sources cited secondhand.
  • The ads analysis is a theoretical model; how ads will affect organic answers is unmeasured.
  • No study links AI visibility to revenue with strong evidence.
  • Two of the studies cited here come from authors who work for AI visibility or monitoring companies.

Frequently asked questions

Will AI search replace Google within five years?

The research cannot say. It shows AI answers on 67% of identical US Google searches in 2025. Clickstream research cited in one study found traditional searches fell by more than 20% after people adopted AI search.

Is AI search a sales channel yet?

Partly. In one 2026 audit, ChatGPT stated a first-person product pick in 79% of product-recommending answers, but different assistants showed very different sources and picks.

Should we move our SEO budget into GEO?

Not wholesale. AI engines run their own web searches, and the review of 45 studies found no GEO technique with proven lasting effects. Keep search strong while building AI visibility.

How often should we revisit our AI search strategy?

At least yearly, with monthly tracking. In one study, roughly 65% of the sources AI engines cited changed from one day to the next.

Sources

Free strategy call

Some questions are easier to answer about your own business.

Bring the one that matters most. On a free 30-minute call we’ll take a first look at it and send you a short written read afterward.

Ask about my businessHow we work