What does Refract do?

Refract builds ChatGPT Ads campaigns from the conversations your customers really have. You start with your website and a sentence about what you sell; the builder takes you through personas, a conversation map, ad groups, context hints and ads, and gives you a bulk-upload file for OpenAI Ads Manager.

Do I need an OpenAI Ads Manager account?

Yes, to run the ads. Refract builds and checks the campaign; you upload the file in Ads Manager, where the ads are reviewed and you set billing with OpenAI. Refract never changes campaigns in your Ads Manager account.

What's included for free?

One campaign with up to 5 ad groups and 25 ads, and 40 AI runs to build it. You can edit everything by hand at any time, and export the campaign when it's ready.

What counts as an AI run?

Each request the builder makes to the AI model, such as drafting personas or writing ads for one ad language. Steps that reuse saved results don't use a run. Your plan and billing page shows how many you've used.

Can I change what the AI writes?

Yes. Every persona, conversation, ad group, hint and ad can be edited, added or removed, and the review step checks claims, character limits and duplicates before you export.

How do clients stay separate?

Each client has its own profile, campaigns and results. People you add to one client only see that client; your team can open every client. Access is checked on every record, not just hidden in the app.

Can clients see their own results?

Yes. Invite them to their client profile as a viewer: each client includes 5 free viewer seats. Client admins can also build campaigns for their own brand and use a team seat.

Where does Conversation Intelligence get its data?

From the results you import from Ads Manager, or sync by connecting an Ads Manager API key to a client, plus each campaign's conversation map and published benchmarks. Refract never reads people's ChatGPT conversations.

How long until Intelligence shows anything useful?

As soon as a campaign has results. Early signals are shown with how sure they are, and they firm up as results build over the first week or two of delivery.

How do upgrades and downgrades work?

Changes apply straight away. Before you switch, Refract shows what Stripe will charge now for the rest of the billing period, and to which card. Moving to a smaller plan leaves a credit for your next invoices.

Open Plan and billing, then Card, invoices and cancelling. Your plan stays until the end of the period you've paid for, then moves to Free. Nothing is deleted.

Do prices include VAT?

Prices are in euros, before VAT. Businesses can add their VAT number at checkout or in Card, invoices and cancelling, and every payment has an invoice you can download there.

Who can see my data, and where is it stored?

Only the people you give access to. Refract runs on Base44, which stores data in the United States under standard data protection terms. The Trust centre, data processing agreement and sub-processor list have the details.

Can I export or delete everything?

Yes. Export a client's data from the Trust centre, delete clients and campaigns yourself, or email us to close your account and delete it all.

← All posts

27 October 2026 · Refract team

Why scale ChatGPT Ads campaigns by conversation theme

One broad ad group is the easiest way to start and the hardest way to learn. Here's why scaled campaigns, with one ad group per conversation theme, work better in ChatGPT.

The quickest way to launch on ChatGPT Ads is one campaign, one ad group, a handful of context hints and three ads. It gets you live. It also tells you very little, and it leaves most of the conversations your customers have without an ad that fits them.

Scaling a campaign doesn't mean spending more. It means building more, narrower ad groups, each around one conversation theme, and covering each market in its own language. Here are four reasons that matters more in ChatGPT than in other channels.

1. The ad group is how you see what works

In ChatGPT Ads you can't see the conversations that triggered your ads, and results are broken down by campaign, ad group, ad and country. There's no search terms report to dig through, so whatever you put in one ad group comes back as one set of numbers.

With one broad ad group, a strong theme and a weak one average each other out. With one ad group per theme, each line in the report answers a question: are people comparing sofa beds with futons clicking? Are first-time buyers converting? That's what you need to move budget and write better ads.

2. Specific ad groups match conversations better

OpenAI's guidance recommends keeping each ad group to one product category, theme or intent, and creating separate ad groups for audiences or use cases that need different messages. An ad group's hints and ads should all describe the same situation. When they do, the ads can answer what the person actually said, instead of a generic pitch that half-fits five situations.

A closer match is also more likely to be clicked, so specific ad groups tend to cost less per click, not just perform better. You can see the effect on a budget in our cost calculator.

3. Conversations happen across the whole decision

People come to ChatGPT at every stage: describing a problem, weighing approaches, checking fit and price, or looking for where to buy. Each stage is a different conversation and needs a different ad, and one ad group can only really speak to one of them.

Multiply the stages by the kinds of customer you serve and the markets you sell in, and even a small business has dozens of conversations worth an ad group. A furniture shop with five types of customer, four stages and three countries already has sixty combinations to consider.

4. Every market needs its own language

People talk to ChatGPT in their own language. An English ad in a Spanish conversation reads as out of place. Scaling across markets means writing hints and ads for each market, not translating one campaign word for word, because people describe the same need differently in each language.

What scaling looks like in practice

More ad groups, each built around one conversation theme, with hints and ads that describe the same situation.
Coverage across the decision, from exploring the need to ready to buy, plus a small catch-all ad group for the conversations you didn't plan for.
Each market in its own language, with hints and ads written for it.
Several ads per ad group, so you learn which message works for each theme.
Budget that follows the themes with the lowest cost per conversion.

The catch: building it by hand

Scaled campaigns take work. Twenty ad groups in three languages means sixty sets of hints and well over a hundred ads, each within Ads Manager's limits of 50 characters for a headline and 100 for a description, and each consistent with its ad group. In a spreadsheet, that's slow and easy to get wrong.

It's the job Refract was built for: it maps the conversations your customers have, turns each theme into an ad group with hints and ads in every market's language, checks them against Ads Manager's limits and gives you one file for bulk upload. For the method itself, whether you build by hand or with Refract, read how to structure a scaled ChatGPT Ads campaign.