Two rival AIs wrote the same ads, word for word

Ask Google’s Gemini for a Burger King campaign and one of the lines it may give you is “Nobody ever gathered around a hot plate.” Ask Grok, which is made by a different company, xAI, with its own engineers, its own data and its own training, and it may give you the same sentence, down to the full stop.

We came across this while measuring something else. We gave both models the same 20 briefs for well-known brands, asked each for ten campaign ideas per brief, and kept the two sets apart. When we compared them, five lines matched word for word.

Several of these are good lines, which is the problem. A good line that two rival AI companies arrive at on their own is a line any brand in the category can get by asking.

Word for word is the rare case

The exact matches are the easy ones to spot. Far more often, the two models landed on the same idea and phrased it a little differently. In the pairs below, the words they share are highlighted.

The more ideas you collect, the more already exist

One batch from each model understates the overlap. So we kept collecting ideas from Grok, in separate sessions, and checked Gemini’s ideas against everything Grok had written so far. Against one batch, about a quarter of Gemini’s ideas were already there. Against five, 58% were.

We ran the same check with Grok against its own earlier batches. After five, 71% of its new ideas were ones it had already written. Moving to a rival company’s AI bought about 13 points of difference.

Why rival AIs think alike

Researchers at the University of Washington, Carnegie Mellon and the Allen Institute for AI found the same pattern on a much larger scale. In one of their tests, they asked 25 models from different companies to “write a metaphor about time,” 50 times each. The answers fell into two clusters: time is a river and, less often, time is a weaver. They called the effect the Artificial Hivemind, and the paper won a best paper award at NeurIPS, the field’s largest research conference, in 2025.

The cause is how these models are made. They learn from largely the same public internet. Then people rate their answers, and each model is tuned toward the kind of answer people rate highly, which tends to be the familiar good one. Every AI company runs a version of that process, so every company’s AI is drawn toward the same likeable middle.

What this means for your brand

Most brands worry that AI will make their work worse. The bigger risk is that it makes their work the same as everyone else’s. If your team, your agency and your competitor’s agency all use AI on its own, they are drawing from one shared pool of likely ideas, whichever company made the tool.

Switching tools changes little, and so does a cleverer prompt. Ask for something bolder and you get the bold version of the same idea. The output changes when what goes in is something a competitor can’t also put in: what people say about your brand and its rivals, your sales history, and the judgment of your best people.

A rule worth bringing to the next creative review: if an AI idea came easily, assume a competitor has it too, and ask what it was built on.