AI Leadership Weekly · Issue #67 · Tuesday 20 January 2026 · 08:00 GMT

Good morning.

Personalisation and advertising have reached the same week’s agenda. Google introduced Personal Intelligence, while OpenAI outlined a test of adverts for some ChatGPT users. Anthropic’s latest Economic Index adds a different perspective: what people appear to be doing with Claude. The dividing lines are important here—personal versus business accounts, a planned test versus a live feature, and sampled usage versus the whole economy.

Gemini adds Personal Intelligence. Google began an opt-in US beta connecting personal Google information.

OpenAI outlines advertising tests. OpenAI announced plans to test advertising for eligible adult US Free and Go users; the announcement preceded the test itself.

Anthropic examines how AI is used. The January Economic Index analysed sampled Claude interactions using measures such as task complexity and autonomy; it was not a workforce census.

https://blog.google/innovation-and-ai/products/gemini-app/personal-intelligence/

Google · 14 January 2026

What happened. Google announced Personal Intelligence on 14 January as an optional US beta for eligible AI Pro and Ultra subscribers. It connects selected personal Google apps to help answer questions with more context and is off by default. Google explicitly excluded Workspace business, enterprise and education accounts from this initial beta.

Why it matters. Our take: Do not turn a consumer demonstration into a workplace instruction without checking the account boundary. More context should have a defined purpose, not become an invitation to connect everything. Ask how someone can inspect the sources, correct a mistaken association and turn the feature off. Convenience and appropriate information use need separate decisions.

What to do. Review one proposed personalised workflow with the information owner. List the sources required and remove those that do not help the task. Confirm that the chosen account and product route are approved. Test whether an answer can be checked without relying on the assistant’s confidence.

https://openai.com/index/our-approach-to-advertising-and-expanding-access/

OpenAI · 16 January 2026

What happened. On 16 January, OpenAI said it planned to test advertising in the US for logged-in adult users on ChatGPT’s Free and Go tiers. It said higher paid tiers would not include ads and outlined principles for separating advertising from answers. These were announced plans and commitments, not evidence of an already completed January rollout.

Why it matters. Our take: The business question is which service your people and customers are actually using. Product funding, account tiers and presentation can matter to the experience. Avoid treating every use of a brand name as the same environment. Keep the organisation’s approved route visible and explain the distinction without claiming a supplier’s stated principles eliminate all concerns.

What to do. Check how your internal guidance names permitted AI tools. Replace broad product labels with the relevant account type and approved configuration. For customer-facing experiments, decide how sponsored material would be identified. Do not assume a planned feature is available locally or build a campaign around an unconfirmed rollout.

https://www.anthropic.com/research/anthropic-economic-index-january-2026-report

Anthropic · 15 January 2026

What happened. Anthropic’s 15 January Economic Index report analysed sampled Claude usage from November 2025. It introduced measures covering task complexity, autonomy and inferred success, among other dimensions, and reported that more complex tasks were less likely to succeed. The sample and model-assisted classifications do not represent every employer or independently verified workplace outcomes.

Why it matters. Our take: A usage chart tells you that something happened, not whether the intended job was completed well. Ask which tasks users return to, where they need help and what changes after review. A smaller number of successful applications may be more valuable than a larger number of unproductive interactions. Keep that possibility open in the scorecard.

What to do. Add a simple outcome question to one team’s pilot log: accepted, corrected, abandoned or escalated. Pair it with task type and approximate review effort. Discuss a few failures with users. Use the conversation to improve the process rather than turning every unsuccessful attempt into a performance judgement.

7 January · OpenAI introduces ChatGPT Health. OpenAI announced a dedicated health-and-wellness experience with staged access. It was designed to support, not replace, care from qualified professionals. Source

5 January · NVIDIA sets out Rubin. NVIDIA announced the Rubin platform and its deployment roadmap. Future infrastructure claims did not establish immediate savings for customers. Source

24 December · Groq licenses technology to NVIDIA. Groq announced a non-exclusive inference-technology licensing agreement with NVIDIA. It was not described as a sale of the entire company. Source

22 December · Atlas receives security hardening. OpenAI described further defences against prompt injection in Atlas, treating malicious webpage instructions as an ongoing security problem. Source

More dated updates from the preceding 30 days.

12 January · Claude introduces Cowork. Anthropic launched Cowork as a research preview for Max subscribers on macOS, allowing work with files in an authorised folder. Source

11 January · Claude expands its healthcare tools. Anthropic announced healthcare and life-sciences connectors and workflow tools. The launch included specialist administrative applications, not evidence of autonomous clinical judgement. Source

11 January · Google proposes a commerce protocol. Google introduced the Universal Commerce Protocol for agent-led shopping, alongside checkout plans and new tools for retailers. Source

8 January · Gmail adds Gemini features. Google announced new Gemini-powered Gmail features, with initial US-English access and different availability across free, paid and testing experiences. Source

Minutes 0–4 · Name the environment. Write down the exact account type, subscription and approved product route. Identify who administers it. Avoid a shorthand product name that hides different settings or makes a personal subscription look like an organisation-managed service.

Minutes 4–8 · Justify the context. List the information needed for the task. Separate essential sources from convenient extras. Decide what should remain disconnected and how a person can verify which source supported an important statement.

Minutes 8–12 · Test a mistaken inference. Use a safe example containing outdated or conflicting context. Ask whether the assistant notices the conflict and whether the user can correct it. Record the source of the misunderstanding rather than simply changing the prompt until it disappears.

Minutes 12–15 · Measure the result. Choose a small outcome measure, such as accepted output or reduced rework. Capture unsuccessful attempts as well as successful ones. Set a review date for both the usefulness of the feature and the appropriateness of its connected information.

Make information access purposeful. The best starting point is an approved environment, a narrow reason to connect data and an outcome the people doing the work can recognise and challenge.

Can we explain why every connected source is necessary for this particular business task?

Brought to you by Data Wave — your AI & Data Team as a Subscription.

Explore Data Wave