89% of marketers say AI search delivered gains for their brand in 2025, yet most cannot quantify the impact with confidence, according to a Business of Apps survey circulated this week.

The figure captures the central tension of the year: AI-driven discovery has moved from experiment to acquisition channel, but the instrumentation around it has not kept pace. Marketers are winning visibility in tools such as ChatGPT, Perplexity, and Google's AI Overviews, then finding the data trail hard to follow back to revenue.

What did the survey actually find?

The headline number is 89%. That is the share of marketers who told Business of Apps they saw measurable gains from AI search activity during 2025. The follow-up problem sits in the second half of the finding: those same respondents said they struggled to measure the impact accurately.

The combination matters. Brands that have shifted budget into AI search now have an industry-wide benchmark for the difficulty of proving it works.

Why is measurement so hard?

Three reasons keep surfacing in practitioner conversations:

• Conversational search splits a single query across multiple model sessions, breaking the keyword-to-page mapping that SEO dashboards rely on.
• AI Overviews often answer the question directly, which suppresses the click that analytics platforms count.
• Referral traffic from large language models tends to arrive without the UTM parameters and referrer strings that downstream tools parse.

The net effect is a familiar one to anyone running paid social: spend goes up, the dashboard says traffic went up, but the link between them is loose.

What changes for the budget conversation?

The finding does not change compliance or platform rules. It changes the budget conversation. Marketing leads asked to defend AI search spend with the same rigor as paid search now have an industry-wide benchmark for the difficulty of doing so.

The practical response from in-house teams has been to build parallel measurement stacks. Common elements:

• Brand-lift studies run quarterly, separate from the performance dashboard.
• A bespoke prompt-share tracker that samples category queries across major models.
• Direct traffic and assisted-conversion models, with AI referrals folded in as a non-last-click source.
• Qualitative win-loss interviews that ask buyers how they first encountered the brand.

None of these are new. What is new is the volume of spend now sitting behind them.

Who is most exposed?

Agencies that built reporting on third-party SEO tools and last-click attribution face the sharpest write-down. Their clients will see AI search traffic inside a general "direct" bucket and ask why the agency cannot break it out.

In-house teams with a mature data engineering function will be better placed. They can wire server-side logging and prompt-monitoring into the same warehouse that holds CRM data, which closes the attribution loop faster than a vendor refresh.

What should operators do this quarter?

• Stop treating AI search as a free channel. The traffic is real, but the reporting gap is also real.
• Stand up a prompt-monitoring routine. Ten category prompts, sampled weekly, costs almost nothing.
• Push vendors on referrer quality. The major LLM products vary in how much traffic metadata they pass.
• Reserve a portion of the AI search budget for measurement. The Business of Apps result suggests the average team under-allocates here.

Where this leaves the industry

The 89% number will be quoted heavily in vendor decks through the end of the year. The measurement problem will not be solved by the next quarterly earnings call. Marketers who can name the AI sources that drove their last twenty closed deals will sit at the table when the 2026 budget gets cut.

The Business of Apps survey is one data point, but the pattern it describes has been visible in earnings calls, agency reviews, and operator Slack threads for two years. AI search is producing outcomes. The numbers behind those outcomes are still being written.