Can we target ads by predicted gender?

We advise against using predictions for ad targeting. They estimate name usage, can be wrong for individuals, and should stay in aggregate analytics or light, optional personalization.

Is analyzing sources and bylines appropriate?

As an aggregate input to editorial review it can be useful, but predictions are imperfect. Where people have stated their gender publicly or to you, use that instead.

How should we handle email-only lists?

Use the email item type in a batch. Expect more unknown results than with full names, and include them in reports rather than dropping them silently.

Gender prediction · Media and publishing

Aggregate reader analytics for media and publishing

Editors and audience teams often ask whether a section, podcast or newsletter reaches a broad readership. Registration forms rarely include demographic questions, and asking can reduce signups. Finowl gender predictions estimate how a name or email local part is typically used, so audience teams can build aggregate reports on newsletter lists or event registrations, labeled clearly as estimates rather than facts about individual readers.

Use the results to inform editorial and marketing choices, not to treat readers differently. Predictions must not decide which articles someone can access, what they pay or which offers they see. They can be wrong for some readers, and they are not how anyone identifies. Mention the processing in your privacy notice and give readers a way to describe themselves if you personalize at all.

Try it in the sandbox About gender

USE CASES

Where it fits in media and publishing

Newsletter audience reports
Estimate the aggregate gender mix of subscribers to each newsletter, helping editors see whether a business, sports or technology newsletter reaches a narrower readership than the site overall.
Source and contributor diversity checks
Estimate the aggregate mix of named sources or bylines across a section over a quarter, as one input for editors reviewing whose voices appear in coverage.
Event and podcast reach
Estimate the overall mix of registrants for live events or podcast listener surveys, helping producers choose panelists and topics that resonate with a wider audience.
Email-based list enrichment
For lists that contain only email addresses, estimate aggregate composition from the local part, accepting that many results will be unknown and reporting them transparently.

HOW TO INTEGRATE

The workflow

1
Export names or email addresses from newsletter lists with opaque ids, separate from advertising profiles.
2
Send batches of up to 50 items to POST /v1/gender/batch, mixing name and email items as needed.
3
Filter by confidence, keep unknown results as a category, and roll up counts by newsletter or section.
4
Publish internal aggregate dashboards and delete per-record predictions after each run.

POST /v1/gender/batch

curl https://api.finowl.co/v1/gender/batch \
  -H "Authorization: Bearer $FINOWL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "items": [
    {
      "id": "nl-sports-001",
      "type": "email",
      "value": "j.moreau@example.com"
    },
    {
      "id": "nl-sports-002",
      "type": "name",
      "value": "Alex",
      "country": "GB"
    },
    {
      "id": "nl-sports-003",
      "type": "name",
      "value": "Fatima",
      "country": "MA"
    }
  ]
}'

Estimates gender for a sample of sports newsletter subscribers to build an aggregate reach report. Results are rolled up and deleted, never used to change content or pricing.

GOOD PRACTICE

Before you ship

Never use predictions to restrict articles, set subscription prices, target ads with significant effects or decide eligibility for offers. They are statistical estimates, not facts about readers.
Keep predictions out of ad-tech and data-sharing pipelines, document the processing in your privacy notice, and retain aggregates rather than per-reader values.
Be transparent in reporting. Label figures as estimates, show the share of unknown results, and avoid presenting predicted data as reader self-identification.

Read the API reference for every field and error code. Each lookup costs 1 credit (3 when it requests AI); see pricing.

A FEW GOOD QUESTIONS

Common questions