Creator data: what the official APIs give you, and what they do not

Anybody assembling creator data hits the same wall in the same order. The official APIs are generous about content and stingy about audience, the numbers you most want are the ones platforms deliberately do not expose, and the scraping that fills the gap sits somewhere between permitted and actionable depending on what you take and from where.

Knowing which data comes from where, before building, saves rebuilding. So does knowing that several of the metrics people scrape are misleading even when correctly collected.

What we'll cover

• What the official APIs actually give you
• The data platforms will not hand over
• Where scraping sits
• The numbers that mislead
• Building it with CodeWords
• Asking the creator, which is underrated
• Personal data, which most of this is
• Designing for a source that will break
• What to collect, and what to compute
• Making it survive
• Limits worth knowing about
• What to build first
• Frequently asked questions

What the official APIs actually give you

The shape is consistent across platforms and the generosity varies a lot.

YouTube's Data API is the most open. Public channel statistics, video metadata, search and playlists, all documented and stable. If your work is YouTube-heavy, most of what you need is available properly.

Most others give you content and identity readily, and audience data grudgingly or not at all. Follower counts are often available; the composition of that following almost never is.

Access tiers matter more than they used to. Several platforms moved to paid API access, and the free tiers are frequently too limited for research work. Budget for this rather than discovering it.

The general rule: anything a platform sells as an advertising or partnership product, it will not give away through a public API. That is the boundary, and it explains most of what is and is not available.

The data platforms will not hand over

Worth listing, because these are exactly the fields people set out to scrape.

Audience demographics. Age, location and gender splits are available to the creator and to advertisers, not to observers.

Reach and impressions. What a post was shown to, as distinct from what it received.

Historical series. Most APIs give a current value, not a trend. If you want a trend, you have to collect it yourself over time, which means starting now.

Deleted or edited history. Gone, and reconstructing it is its own problem.

Anything about the audience as people. Deliberately, and increasingly so.

For the first two the honest routes are asking the creator, or using the platform's own partnership tools. There is no clever technical answer.

Where scraping sits

Worth being accurate rather than either alarmist or dismissive.

Platform terms generally prohibit automated collection. Breaching them is a contractual matter and the practical consequence is blocking or account termination rather than anything more serious.

Law varies by jurisdiction and by what you take. Publicly visible information, accessed without circumventing a technical restriction, sits differently from anything behind a login or an anti-bot measure. Circumvention and authentication are where the position gets materially worse.

Data protection applies regardless. Creator profiles are personal data in most jurisdictions, and the fact that information is public does not remove the obligations that attach to collecting and storing it.

So the sensible order is official API first, then the platform's own partnership tooling, then asking, and only then considering scraping public pages for the residue. Where you do scrape, take the minimum, identify yourself honestly, respect rate limits, and do not circumvent anything.

The numbers that mislead

Frequently overlooked, and it undermines a lot of collected data.

Follower count is the weakest signal available and the easiest to inflate. Treating it as the primary sort order gets you a list ordered by who has bought the most followers.

Engagement rate depends on its denominator. Engagements over followers and engagements over reach give very different answers, and the second is the meaningful one and the one you cannot see.

Averages hide the distribution. One viral post lifts an average enormously while telling you little about typical performance. The median is more useful and almost nobody computes it.

Recency matters more than volume. A creator whose last six months are flat looks fine on lifetime totals.

Cross-platform comparison is mostly invalid, since a like means something different on each.

A practical response: compute the median rather than the mean, weight recent activity, and treat follower count as a filter rather than a ranking.

Building it with CodeWords

CodeWords connects to more than 3,000 integrations, and the connection is made once and reused.

1. Open CodeWords and start a new automation.
2. Describe what should happen in plain language to Cody, the automation builder: which creators to track, which fields matter, and how often to refresh.
3. Authorize the connections to the platform APIs you have access to and wherever the data should land.
4. Describe the exceptions: a source that returns nothing, a rate limit, a creator whose account has gone.
5. Store a timestamped snapshot on every collection rather than overwriting, since the series is the part you cannot get retrospectively.

You describe the outcome; Cody builds it, connects it, and deploys it. The free plan covers light use, with Pro at $39 per month and Business at $100 per month as usage grows; details are on the pricing page.

Asking the creator, which is underrated

For the fields that matter most, this is the route that works.

Creators can see their own demographics, reach and historical performance, and will usually share a screenshot or an export for a serious enquiry. Many have a media kit already. It takes a message.

Three advantages beyond availability. The data is the platform's own rather than your estimate. You learn something from how they respond, since a creator who provides a clear export is a different proposition from one who cannot. And you have made contact, which was the point of the research.

The obvious caveat is that self-reported figures can be selective, so ask for the platform's own screens rather than a summary. Use the collected data to shortlist and the direct request to verify, rather than trying to compute your way to something the platform will not expose.

Personal data, which most of this is

A creator profile is personal data, and a database of them is a database of personal data regardless of how public each item was.

The obligations follow: have a lawful basis for holding it, be able to say what you hold and why, apply a retention period rather than accumulating indefinitely, and be able to respond if somebody asks what you have about them. Public availability is not consent and does not exempt you from any of this.

Two practical consequences. Collect the fields you will actually use rather than everything available, because every extra field is more to justify and more to delete. And keep the data somewhere you can query and purge, rather than scattered across exports.

Designing for a source that will break

Whatever you build on will change, and scraped sources change without notice.

Isolate each source behind its own component, so one breaking does not stop everything.

Validate what comes back. A follower count of zero is more likely a broken parse than a real value, and a pipeline that stores it without complaint quietly corrupts the dataset.

Alert on shape changes, not just on errors. Silent success returning nothing useful is the failure that goes unnoticed for weeks.

Keep raw responses for a while, so a parsing bug can be corrected retrospectively rather than requiring recollection.

Expect access tiers to change, and design so a source can be swapped rather than being woven throughout.

What to collect, and what to compute

Collect the primitives and derive the rest, because derived values change definition and primitives do not.

Store per observation: the timestamp, the identity, the follower or subscriber count, the post identifiers with their own timestamps, and the raw engagement counts. That is enough to compute anything later.

Compute rather than store: engagement rates, medians, growth rates, posting cadence, and any score you invent. When you change your mind about the formula, and you will, recomputing from primitives is an afternoon while recollecting is not possible at all.

Making it survive

Official API first, every time.

Snapshot with timestamps, never overwrite.

Store primitives and compute the rest.

Validate values before storing them.

Apply a retention period, since this is personal data.

Limits worth knowing about

Audience demographics and reach are not available to observers, by design.

Follower count is the weakest signal and the easiest to inflate.

Platform terms generally prohibit automated collection, with blocking as the usual consequence.

Public availability does not remove data protection obligations.

Historical series cannot be obtained retrospectively, which is the argument for starting collection before you need it.

What to build first

A timestamped snapshot of the primitives for a small set of creators, from official APIs only, running on a schedule. Not a score, not a dashboard, not a scraper. The thing you cannot get later is the series, and every week you delay is a week missing from it.

Two habits make the difference. Store raw counts and compute every derived figure, because you will change your definition of engagement rate and recomputing is an afternoon while recollecting is impossible. And ask creators directly for the fields the platforms withhold, since a message gets you better data than any amount of inference and it starts the conversation you wanted anyway.

Frequently asked questions

Which platform APIs give the most creator data?

YouTube's Data API is the most open, with public channel statistics, video metadata, search and playlists. Most others provide content and identity readily and audience data sparsely or not at all.

Why can I not get audience demographics?

Because platforms sell that as an advertising and partnership product. It is available to the creator and to advertisers, and deliberately not to observers.

Is scraping creator data legal?

It depends on what you take and from where. Platform terms generally prohibit it, and the legal position worsens considerably where you circumvent a technical restriction or access anything behind a login. Data protection applies regardless.

Does public data avoid data protection obligations?

No. A creator profile is personal data, and public availability is not consent. You still need a lawful basis, a retention period and the ability to answer a request about what you hold.

Why is follower count a poor metric?

It is the easiest number to inflate and the least connected to outcomes. Using it as a ranking produces a list ordered by who bought the most followers.

What is wrong with engagement rate?

The denominator. Engagements over followers and engagements over reach give very different answers, and the meaningful one uses reach, which you cannot see as an observer.

Should I use the mean or the median?

The median, weighted towards recent activity. One viral post distorts an average substantially while telling you little about typical performance.

How do I get the data platforms withhold?

Ask the creator. They can see their own demographics and reach, most have a media kit, and how they respond tells you something useful in itself.

What should I store?

Timestamped primitives: identity, follower count, post identifiers and raw engagement counts. Compute rates, medians and scores from those, since definitions change and primitives do not.

How do I handle a source that breaks?

Isolate each source, validate returned values rather than storing them blindly, alert on shape changes as well as errors, and keep raw responses so a parsing bug can be fixed retrospectively.

Related reading

• Where AI actually helps a channel
• Scraping prices, and why it is adversarial
• Which Google credential you need