Trust centre
Your clients' data stays theirs
Refract works out its intelligence itself, from the data you bring. Here's exactly what we keep, who can see it and the few services that handle it.
Your data works only for youEach client's intelligence is worked out from that client's own campaigns and results. Nothing is pooled unless the workspace owner switches on benchmarks.
No data brokersRefract doesn't buy, sell or rent data, and doesn't send your results to analytics, ad-tech or data-enrichment services.
Every number is explainedIntelligence labels each number: Observed (your results), Modelled (your conversation map), Estimated (Refract's forecasts), Benchmark or Public data. “Why?” shows the working.
Yours to take or deleteExport everything stored for a client as one file at any time. Deleting a client deletes its intelligence with it.
What we store
Only what the product needs
Your accountName, email address and sign-in details.
Workspace and clientsClient names, websites and the descriptions you write, plus who on your team can open each client.
Campaign buildsPersonas, the conversation map, ad groups, context hints, ads and images you upload.
Results you importDaily impressions, clicks, spend and conversions by ad group, ad and country, from Ads Manager reports you import or sync. These are totals: nothing about the people who saw the ads.
Ads Manager connectionOptional, per client: the account name and the API key you paste, stored encrypted and never shown again. Used only to read that client's reports. Deleted when you disconnect or delete the client.
BenchmarksOnly if your workspace opts in: one CTR per funnel stage and market from your real results, combined with at least 4 other workspaces before anyone sees it.
IntelligenceSignals, suggestions and the analysis worked out from the rows above.
BillingYour plan and a Stripe customer reference. Card details are held by Stripe and never reach Refract.
Refract never reads anyone's ChatGPT conversations. Conversation maps are simulated by AI from your brief, and results come from the Ads Manager reports you choose to import.
Who can see it
Locked to your team and each client
Your teamOwners, admins and editors see every client in the workspace. Viewers can look but not change anything.
Each client's peoplePeople you invite to a client see that client only: never your other clients, their results or their intelligence.
Checked on every rowAccess is enforced by the database on every record, and every change goes through server checks. It doesn't rely on hiding things in the app.
Where it goes
The few services that handle it
| Service | What for | What it receives |
|---|---|---|
| Base44 | Hosting, database, sign-in, emails (sign-in codes, invites) and basic usage statistics for the app. | Everything listed above, stored on your behalf, plus which pages are opened and for how long. |
| Base44's AI service | The AI steps in the campaign builder and the location explorer: page summary, personas, conversation map, hints and ads. | Your campaign brief, the text of the public pages you ask it to scan, and the builder content for that step. Never your imported results. |
| Stripe | Payments, invoices and the billing portal. | Billing name, email, address and payment details, entered on Stripe's own page. |
| OpenAI Ads API | Only for clients you connect: reading that ad account's delivery and conversion reports. Refract never changes campaigns through it. | The API key you gave and the report request (dates and campaign). Nothing about you, your other clients or your Refract data. |
- Conversation Intelligence uses no outside data service and no AI. Signals, suggestions, theme health, forecasts and benchmarks are calculated by Refract from your rows, the same way every time. The only outside call is reading your reports from OpenAI's Ads API, and only for clients you connect.
- When you scan a website, Refract's server reads that public page directly. Nothing else is sent to the site.
- Base44's AI service runs on models from OpenAI and Anthropic. The full list of sub-processors, including the providers Base44 uses, is on the sub-processors page, and our data processing agreement covers the data you add about your clients.
- No advertising trackers and no outside analytics services. Fonts are served by Refract itself, so opening Refract doesn't tell any other company you're here.
Public data
Open datasets we processed ourselves
OpenAI Signals v2.0OpenAI's published shares of ChatGPT messages by country, topic, age group and typical first-name gender, plus a ranking of countries by messages per person. Refract uses it for ChatGPT usage and “Who's asking” in the Explorer, and the age mix in the persona step. We averaged the latest three months (Apr 2026 – Jun 2026) and kept the markets Refract supports.By Aaron Chatterji, Thomas Cunningham, David J. Deming, Zoe Hitzig, Drew Johnston, Alex Martin Richmond, Christopher Ong, Carl Yan Shan and Kevin Wadman. Licensed CC BY 4.0. Source.
WildChat-4.8MAbout 3.2 million conversations people chose to share with a research chatbot (Apr 2023 – Jul 2025). Refract counted how often first messages mention each product category, per country and language, for “Interest in your category” in the Explorer. We read only the message text, language and country, kept counts only, counted repeated scripted prompts at most 3 times, and dropped any count under 30.By the Allen Institute for AI. Licensed ODC-BY 1.0. Source.
Both were processed once on Refract's own systems and ship with the app as small tables of shares. Using these features doesn't contact OpenAI, the Allen Institute or anyone else, and nothing about your campaigns goes into them.
Questions about your data?Ask for a copy of what we hold, a correction or deletion, or details for your own privacy notice.
Contact us