The five targeting levers
| Lever | What it does | Limit |
|---|---|---|
| Context hints | Free-text descriptions of the topics your ad should match | Up to 2,000 per ad group |
| Location | Countries and location IDs, including exclusions | Up to 2,500 entries |
| Platforms | Android app, Android web, desktop web, iOS app, iOS web | Chosen per campaign |
| Custom audiences | Hashed email or phone lists | Inclusion needs 25,000 matched users; exclusion has no minimum |
| Bidding | Views, clicks or conversions | One objective per campaign, fixed after creation |
What you can't target
- Companies by name, domain or IP address.
- Job titles, functions or seniority.
- Lookalike audiences or past website visitors.
- Anything a user typed. Advertisers never see conversations.
How B2B teams use each lever
Context hints. Treat each ad group as one buyer question at one stage. "Month-end close software for a mid-size logistics company" is a better hint than "accounting software", because only buyers like yours describe that problem. Our guide to writing context hints has examples.
Location. Match your sales territories. To restrict delivery to the US, OpenAI says you must include the United States explicitly; exclusions alone don't restrict delivery.
Platforms. Split desktop web into its own campaign. Work research happens at desks, and company identification works better on office networks.
Custom audiences. Most B2B customer lists are far below 25,000 matched users, so use them to exclude customers, staff and open deals. Custom audiences aren't available for campaigns in the EEA or Switzerland.
Bidding. Start on clicks. Move to conversion bidding once demo requests are tracked. Which bidding to use.
The workaround for account targeting
You can't choose the company before the click, so choose the problem and name the company after it. Company identification on your website matches visits from business networks to company records. How that works.
A worked example
A fictional close-management software company sells to finance teams at 100–1,000 person companies in the US and UK. Its targeting plan:
| Campaign | Location | Platform | Ad groups (one buyer question each) |
|---|---|---|---|
| US decision | United States | Desktop web | "Close software for NetSuite users", "Close tools for multi-entity groups" |
| US problem | United States | All | "How to shorten month-end close", "Reconciliation takes too long" |
| UK decision | United Kingdom | Desktop web | "Month-end close software for UK groups", "Close tools that handle VAT" |
Customers, staff and open deals are excluded with one hashed list across all three campaigns. Each ad group has 20–40 context hints: different phrasings of the same question, with company size, ERP and industry variations.
How targeting interacts with budget
Narrow ad groups spend more slowly. That's expected: a hint that only your buyers would match reaches fewer people. If an ad group spends less than half its share for two weeks, widen it with more phrasings of the same question before you widen the question itself.
Questions
Can I target by industry?
Not directly. Put the industry into the context hint: "inventory software for a food distributor" reaches a different conversation from "inventory software". The industry is part of the question.
Can I target people who visited my website?
No. ChatGPT ads don't offer retargeting or lookalike audiences. Use other channels to retarget identified companies.
How many context hints should an ad group have?
Enough to cover the ways buyers phrase one question. In practice 20–60 is plenty; the limit of 2,000 is there for large advertisers, not a target.
Start here: ChatGPT ads for B2B. More articles in ChatGPT ads.