Organic search is shrinking, and AI is only part of the reason

A lot of marketing teams are looking at the same chart right now: organic traffic sliding, and someone in the room saying "AI did this." This week's news says the story is messier, and more useful.

What happened

A two-year study of search traffic found that organic visits fell 13% in the US and 24% in France. Publishers lost about 42% in both markets, while smaller sites stayed roughly flat. Paid search traffic in the US grew 21% over the same period, as bigger ad blocks pushed organic results below the fold. AI referrals were just 0.5% of total visits.

So the loss is real, but the study points first at the ads above the results, not at ChatGPT.

Meanwhile the platforms keep moving. Meta's feeds on Facebook, Instagram and Threads now run more on AI predictions of what people want than on the accounts they follow. Google updated its privacy policy so Search can use connected Gmail, Drive and Photos content to personalize ads. And Pennsylvania signed an AI Deceptive Advertising Act: if AI-generated content creates a false impression about a product, brands must disclose it at the first point of exposure, whatever their size. California and New York are working on their own rules.

What it means if you lead marketing

Three things stand out to me.

First, "followers" and "rankings" are becoming weaker assets. When a feed is picked by a model and the top of the search page is sold, your reach depends on whether the content earns attention on its own, again and again. That favors teams that publish steadily over teams that wait for a big launch.

Second, attribution is getting worse at the exact moment budgets are getting tighter. One piece this week made the point well: in a zero-click world, the tutorial or video that earned someone's trust two years ago never gets credit for the sale. Your dashboard undercounts your best work.

Third, the rules on AI content are arriving state by state. Because nobody yet knows exactly what counts as a "false impression," the cautious reading is to disclose AI use more broadly, not less.

What I would do

• Talk to ten recent customers this month. Ask what convinced them and where they first heard of you. Then use AI to find the patterns across those conversations. It will tell you more than last-click ever will.
• Give the ad algorithm a mandate before a budget. A post this week argued for spending caps on brand searches and existing customers, clear approval rules, and holdout tests to check what the platform claims. I agree. Let the machine bid, but decide the boundaries yourself.
• Write a one-page AI disclosure policy now. Where you use generated images, voices or copy, say so at first exposure. It is cheaper to set the habit early than to retrofit it under a new law.
• Put your team's AI files under version control. One B2B content team moved their shared voice files, positioning and AI skills to GitHub, without developer help, so every person's Claude reads the same source and an off-brand result can be traced and rolled back. Once you share more than a handful of these files, that discipline matters.

Worth trying

Brave Search URL submission. Claude's web search runs on the Brave Search API, and Brave lets you submit pages to be re-fetched. If you care about showing up in AI answers, submitting your key pages after each update is worth adding to your publishing checklist. It is a small, honest step toward visibility in AI search, with none of the gimmicks.

The common thread: the channels are changing faster than the dashboards. The teams that keep a person close to the customer, and set the rules before the algorithms spend, will read this shift correctly.