The Maze: AI has become marketing's busiest employee and one of its least convincing salespeople. In an NP Digital survey of 1,903 marketers, 91% said AI had not increased revenue. Only 9% said it had. The useful part sits inside that small minority: 89% of the revenue-positive group credited traffic from large language models.
That is a different payoff from the one most AI budgets promise. The common pitch is production: more copy, faster creative, cheaper variants. Yet just 4% of the successful subgroup credited better-converting AI creative, while 2% pointed to AI content driving traffic or sales. The commercial winner was distribution. AI made money when it delivered a customer, not another asset.
The denominator matters. The 89% is not 89% of all marketers. It is 89% of the 9% who reported a gain. Multiplying the rounded figures gives an indicative 8% of the total sample, but that is not a measured survey result. The source also does not publish geography, respondent roles, field dates, sampling frame, industry mix or uncertainty. Treat the result as a sharp operating signal, not a universal ROI benchmark.
Still, transaction data points in the same direction. Adobe Digital Insights analysed more than 1 trillion US retail visits and found AI-sourced traffic up 393% year over year in the first quarter of 2026. In March, those referrals converted 42% better than non-AI traffic, reversing a 38% deficit a year earlier. Visitors from AI also spent 48% longer on site and viewed 13% more pages.
The catch is that earning this traffic requires infrastructure, not prompt theatre. Adobe found average machine-readability scores of 75% for homepages, 74% for category pages and only 66% for product pages. If an assistant cannot parse availability, attributes, reviews, policies and proof, the brand may never enter the answer—however efficiently its marketing team generates campaign copy.
Separate production from acquisition. Track AI time savings and AI-attributed demand as different investment cases. Faster output is an efficiency benefit; referred customers are a growth benefit.
Make the catalogue legible. Product facts, structured data, merchant feeds, policies and third-party evidence now form a distribution layer for assistants as well as search engines.
Measure direct and assisted discovery. Tag referrals where possible, monitor source and medium, ask customers how they found the brand and examine branded-search lift. A later search or untagged visit can hide the earlier AI influence.
Optimize beyond the citation. Visibility only opens the door. Landing-page relevance, trust, price, availability and checkout determine whether a referral becomes revenue.
Attribution matters. Google Analytics records the observable source or referrer; visits without a clear one can land in direct traffic. A shopper can discover a brand in an assistant, later search its name and give the visible session credit to search. That makes the 91% partly a measurement challenge. Hidden influence is not an excuse for missing commercial evidence.
The management question is therefore not “How much content did AI make?” It is “Where did AI move a customer?” Teams that cannot answer the second question have automated the supply of marketing without building demand. That is productivity. It is not yet growth.
