ContextHint
Practical guide

How Many Context Hints Per Ad Group?

OpenAI does not recommend a universal number. Use enough closely related hints to describe one buyer need—and start a new ad group when the product, message, or landing page needs to change.

The one rule

Would the same ad make sense for every hint?

If yes, the hints probably belong together. If one of them needs a different promise, proof point, or destination, give it its own ad group.

Make the decision

Keep related situations together. Split different jobs.

Keep them together
  • The same product solves every situation.
  • The buyer has the same underlying need.
  • The same ad promise makes sense for every hint.
  • Every click can go to the same landing page.
Start a new ad group
  • The product or offer changes.
  • One audience needs different proof or positioning.
  • The conversation moves to a different decision stage.
  • A hint needs different creative or a different destination.
Worked example

One invoicing product, three related situations.

These hints describe different moments for the same kind of buyer. One product, one core problem, and one landing page can serve all three.

  1. 01

    A freelancer comparing invoicing tools because chasing late payments takes too much time.

  2. 02

    A consultant billing international clients who needs currencies and automatic reminders.

  3. 03

    A newly self-employed designer setting up a professional billing workflow.

30-second check

Ask these four questions before you add another hint.

  1. 1

    Is it for the same product?

  2. 2

    Does it solve the same underlying need?

  3. 3

    Can the same ad answer the conversation honestly?

  4. 4

    Can the same landing page finish the job?

Four yeses: keep it in the group. One meaningful no: split it.

Planning several markets or ad groups? Use the MCP research workflow.

Sources and limits. Reviewed July 22, 2026.

OpenAI says to keep each ad group focused and separate meaningfully different products, audiences, and use cases. Its bulk-upload checklist represents context hints as a JSON list, but it does not prescribe a universal number. Context hints guide matching; they do not guarantee delivery.