AI answers file Google DeepMind under frontier labs, but rarely say what it is credited for

This snapshot is from 2026-09-29 and covers benchmark questions that never named a brand. Google DeepMind was named in 691 of 1,000 answers counted. The count by assistant was Grok 229 of 250, ChatGPT 187, Claude 140 and Perplexity 135. No Gemini count is reported in the data, so the total rests on the other four assistants. Visibility by question intent was highest on innovation (90%), safety (80%) and leadership (79%). It was lowest on questions about who to avoid (43%), service (55%) and data privacy (57%).

The typical framing is a place on the shortlist of frontier developers. ChatGPT, asked who has the most powerful models, said the frontier "appears to be split among OpenAI, Anthropic, and Google DeepMind." Perplexity, asked which foundation-model companies lead, described it as a top-tier developer backed by Google's infrastructure. Claude's answer about breakthroughs tied it to "the Gemini family." In several answers DeepMind blends into Google, Alphabet and Gemini. ChatGPT's answer about the most users worldwide credited "Google/Alphabet" and the Gemini app.

Specific credit is thin. Of the credit tags recorded for the brand, 33 were "Not Substantive," 2 were Safety & Compliance and 1 was Financial Performance. DeepMind is mostly named as a member of a list rather than praised or criticised for something distinct.

Where it appears in risk contexts, the answers attributed the points to outside sources. Claude, asked which companies are most responsible about safety, said several labs including Google DeepMind reversed earlier bans on military applications. Grok's answer on legal risk cited a benchmark with Gemini at low compliance with EU law. Claude's answer on image and chart understanding named DeepMind but led with Anthropic's models and gave Gemini 3 Pro as a general multimodal pick.

PeakMetrics' news index shows 43,848 mentions over the last 30 days, with 5,480 carrying the brand in the headline. The sampled headlines are dominated by Gemini as a product. They include a Google for Developers page on Gemini-powered solutions, a report that Gemini hacked three companies in a cybersecurity test, and a Gemini notebook update adding an audio recorder and voice chat. One headline is about DeepMind alumni building tools to speed up fusion power for the grid.

Coverage is heavy in absolute terms but smaller than that of the two labs it is most often grouped with. In this data OpenAI has 155,993 news mentions and Anthropic 132,852, against 43,848 for DeepMind. All three appear in a similar share of answers.

The press covers Gemini the product: developer tools, feature updates and a security-testing incident. The answers place Google DeepMind as a research lab in the frontier group and mostly avoid product news. The security story appeared in the press sample, and the answers did not mention it in the text provided here.

Visibility does not translate into a distinct reputation. The answers name DeepMind widely, but the credit tags show little attribution of specific strengths. Its narrative is also shared with Google and Gemini, so the answers do not consistently separate the lab from its parent.

Check how each assistant separates or merges DeepMind, Google and Gemini. This snapshot shows them used interchangeably, so a comms team should not assume a claim about one will be attributed to it.

The weaker intents are service, data privacy and who-to-avoid questions. Answers to those questions leaned on third-party benchmarks, policy reports and safety assessments. Those are the sources that could shape how the brand is described there.

These are single live answers from one day. Re-run the same benchmark questions before reading any of this as a pattern, and treat the missing Gemini count as a gap in the data.