AGI for Marketing

Marketing is where AI adoption is deepest and where the measurement is most confusing, because the metrics that improve and the metrics that matter are not the same ones.

Clicks up, conversions down

Industry benchmark analysis across tens of thousands of ad variations found AI-generated creative outperforming human creative on click-through rate on Meta, roughly 1.08% against 0.96%. Then the picture inverts as consideration rises:

• Under $100 average order value: conversion at parity
• $100 to $500: AI creative converts 8% worse
• Above $500: 14% worse
• B2B qualified opportunities: 18% worse

These are industry benchmarks rather than peer-reviewed work, and the methodology is only partly disclosed, so hold them loosely. But the shape is consistent with everything else in this brief: AI produces surface polish efficiently and substance less well, and the higher the stakes of the decision, the more the substance is what closes it.

If your funnel is judged on CTR, AI creative looks like a clear win. If it is judged on revenue per impression at high AOV, it may be costing you.

The perception gap is widening, not closing

The IAB surveyed both sides of this and found the gap growing. 82% of ad executives believe young consumers feel positive about AI-generated ads. 45% of those consumers actually do. That is 37 points, up from 32 points in 2024.

The innovation halo is going the same way. Consumers calling AI-using brands innovative fell from 30% to 23% between 2024 and 2026. Over the same period advertiser belief that AI signals innovation rose from 40% to 49%. Both lines moved, in opposite directions.

One important counterweight, because it cuts against the doom reading: 73% of consumers said knowing an ad was AI-made would increase or make no difference to their purchase likelihood. Trust and transaction are different measures, and conflating them produces bad strategy.

What disclosure actually does

The academic evidence here is genuinely split, and the split is useful. One 2026 study found disclosure raises perceived novelty while lowering perceived authenticity, two opposing pathways running at once.

Another, with 370 participants, measured trust across three conditions and found a steep gradient: no AI information scored 4.18, "AI-assisted" scored 3.56, and "AI-generated" scored 2.30. All differences significant, all mediated by perceived authenticity.

That middle tier is the practical finding. The penalty for AI-assisted work is a fraction of the penalty for AI-generated work. Disclosing human authorship with AI assistance is a materially different proposition from disclosing AI authorship, and most teams are choosing between disclosure and silence without noticing there is a third option that is both honest and far less costly.

The traffic problem is bigger than the creative problem

This is the change most likely to affect a marketing plan in the next year, and the numbers are not marginal.

68.01% of US Google searches ended without a click between January and April 2026, up from 60.45% in 2024. Where AI Overviews appear, click-through drops by nearly 60%.

Publisher-side measurement across thousands of sites over two years: small publishers lost 60% of search traffic, medium-sized 47%, large 22%. Google Search fell from around 8.5% of global page views in early 2024 to 5.8% by February 2026.

And the replacement is not arriving. ChatGPT sends under 1% of all publisher page views. Perplexity sends 0.002%. Cloudflare's May 2026 measurement put all AI engines combined at 0.29% of search referrals against Google's 87.63%.

The referrals that do arrive convert better, roughly 7% on transactional sites against Google's 5%, with double the time on site. High intent, tiny volume. That is worth optimising for and it is not a traffic strategy.

A caution on the statistics you will be quoted

Estimates of how often AI Overviews appear range from 15.7% to 60.3% depending on which keyword panel was sampled. That is a fourfold spread. Anyone citing one number without naming the panel is not giving you information you can plan against.

Compliance stopped being optional

The EU AI Act's Article 50 transparency obligations came into application on 2 August 2026. In practice: systems must be designed so people are told they are interacting with AI, AI-generated content must carry machine-readable marking, and disclosure must happen at the moment of contact rather than in fine print. Deepfake labelling applies even where no deception was intended. The editorial-review exemption is narrow and requires substantive review by qualified people, with attributable editorial responsibility. Penalties reach €15 million or 3% of worldwide annual turnover.

Legal analysis of the guidance names AI-generated testimonials and product visuals specifically as marketing exposure. Worth noting for the US: there is no FTC rule specifically mandating AI content disclosure in advertising, despite a great deal of content claiming otherwise. Enforcement runs through existing deception authority and the Endorsement Guides.

What to do with this

Four things the evidence supports.

• Measure at the bottom of the funnel, not the top. CTR is the metric AI creative most reliably improves and the one least connected to revenue at high consideration.
• Stop planning around search referral volume. Two-thirds of searches end without a click and AI engines are not replacing what was lost. Plan for owned channels, citation presence and direct relationships.
• Pick the AI-assisted position and say it plainly. The trust penalty is far smaller than for AI-generated, and it is defensible because it is true.
• Get the EU compliance done. It is in force, the marking requirement is technical work with a lead time, and the penalty is turnover-scaled.