Platform and disclosure signals
Native partnership flags and disclosure language such as paid partnership, ad, sponsored, gifted, or partner provide direct commercial evidence.
Sponsorship detection methodology
How Streamforge Identifies Sponsored Creator Content
Streamforge combines platform-native signals, disclosure language, brand context, promotional evidence, and relationship classification to identify likely sponsorships without treating every brand mention as paid.
Nick LombardiCo-Founder & CTO, StreamforgeLast reviewed
- Multi-signal evidence
- No single hashtag is required
- Four relationship types
- Sponsorship, affiliate, first-party, organic
- Confidence floor
- Qualified evidence must clear a threshold
The short version
Detection starts with the relationship, not the word sponsored
Creator content can mention a brand for many reasons. The method evaluates native paid-partnership flags, multilingual disclosure language, named third-party brands, codes and links, gifting, travel, events, and broader context—then classifies the commercial relationship before deciding whether the content counts as sponsored.
The signal groups the model evaluates
Brand and promotion signals
Named brands, promotional codes, affiliate links, calls to action, gifting, and paid-event context strengthen or change the classification.
Relationship context
The model separates third-party brand sponsorship from affiliate activity, a creator promoting their own business, and an organic mention.
The classification pipeline
- 01
Extract candidate evidence
Find native platform markers, disclosures, brands, links, codes, campaign language, and commercial context.
- 02
Identify the relationship
Classify the evidence as brand sponsorship, affiliate, first-party, or organic activity.
- 03
Evaluate confidence
Test whether the combined evidence is strong enough to clear the sponsorship confidence floor.
- 04
Store the qualified result
Count only qualifying brand sponsorships in sponsorship intelligence and preserve the supporting context.
Why counting disclosures undercounts sponsorship
The simplest way to build a sponsorship dataset is to search for disclosure markers and count the hits: the #ad and #sponsored hashtags, the "includes paid promotion" notice, the "Paid partnership with" label. It is easy to build and easy to explain, and it produces a number that is wrong in a predictable direction. A disclosure-only count does not measure how often creators are paid. It measures how often creators disclose, which is a different and consistently smaller quantity.
Three things drive the gap. Disclosure compliance is uneven: the obligation is real under the FTC Endorsement Guides and their equivalents elsewhere, but enforcement is sparse and understanding varies widely across a creator population that includes a great many people who have never read a guideline. Native platform labels are set by the person uploading, so an unset toggle produces an unlabelled sponsored post, and the label is not always rendered identically in every surface, embed, or region where the content is seen. And a large share of real disclosure never enters the text layer at all: it is spoken in the first fifteen seconds, burned into a lower third, or written in a language the collector was not looking for.
So the method treats a disclosure as strong evidence rather than as the definition. Native partnership flags, disclosure vocabulary across multiple languages, named third-party brands, promotional codes, affiliate links, campaign phrasing, gifting and travel context, and event attendance are each partial evidence. The classification comes from the combination. A creator who says nothing but reads a scripted product benefit, shows a code, and links a tracked URL is more likely sponsored than a creator who writes #ad under a photo of their own dog.
Four commercial relationships, not one sponsored flag
A boolean sponsored field collapses four situations that behave differently and carry different consequences for whoever reads the data. A brand sponsorship is a third party paying for placement. An affiliate arrangement pays the creator a share of what their audience buys, with no guaranteed fee and often no relationship with the brand at all. First-party promotion is a creator selling their own merchandise, course, game, or company. An organic mention is somebody talking about a product because they like it.
Conflating these produces specific, repeatable errors. The most common is competitive intelligence that overstates a rival brand's spend, because every affiliate link carrying that brand's name is counted as a paid campaign. The second most common is a creator who looks heavily sponsored and is not: their apparent commercial load is their own store, which tells you about their business model rather than about their availability or their audience's tolerance for advertising. The third is treating an organic mention as a prior relationship and opening an outreach email with a thank-you for a partnership that never existed.
The classification is therefore the primary output and the sponsored/not-sponsored answer is derived from it. Only qualifying third-party brand sponsorships are counted in sponsorship intelligence. Affiliate activity, first-party promotion, and organic mentions are preserved and labelled rather than discarded, because each of them answers a real question — how a creator monetises, what they will endorse unpaid, and which categories they already work in.
What a public-evidence classification cannot establish
Everything here is inferred from what was published. That sets hard limits which are worth stating plainly, because the failure mode for this kind of data is not a wrong answer — it is a confident answer used for a decision it cannot support.
The method does not observe fees. A high-confidence sponsorship says a commercial relationship is evident; it says nothing about what was paid, whether it was paid in cash or product, or whether the creator was on a retainer covering several posts. It does not observe contract terms: exclusivity windows, usage rights, category restrictions, and renewal options leave no public trace, and a competitor's absence from a creator's recent content is not evidence that an exclusivity period has lapsed. It cannot, by construction, find undisclosed arrangements that also left no promotional signal — a payment for silence, or for a competitor's omission, is invisible to any method built on published content.
It also cannot reliably separate gifting from payment. A creator sent an unsolicited product who chooses to feature it has a disclosable material connection and often discloses it identically to a paid integration. Where that distinction matters — most often in rate benchmarking, where free product inflates the apparent volume of paid work — the honest move is to treat the signal as a partnership of unknown value and confirm the terms with the creator or their representative during outreach. The classification is a research input and a market signal. It is not legal proof of a contract and should not be presented as one.
Common questions
Can the model detect sponsorships in multiple languages?
Disclosure language is evaluated across multiple languages alongside other platform, brand, and promotional signals.
Does a promo code always mean sponsorship?
No. It may indicate affiliate or other commercial activity. The relationship classification and surrounding evidence determine the result.
Can Streamforge see private sponsorship contracts?
No. The method classifies observable public evidence; it does not claim access to contracts, fees, or undisclosed terms.
Why not just count posts tagged #ad?
Because that measures disclosure compliance rather than sponsorship. Disclosure rates vary by creator, platform, region, and language, and a great deal of genuine disclosure happens in speech or on-screen text rather than in the caption, so a hashtag count has a floor set by how many creators tag correctly, not by how many were paid.
Does Streamforge know what a creator was paid for a sponsorship?
No. The method classifies observable public evidence of a commercial relationship. Fees, contract terms, exclusivity windows, and usage rights are private and are not inferred from public content. Any rate figure should come from the creator, their representative, or your own past deals.
How is gifted content treated?
Gifting is captured as commercial context, but it is not treated as equivalent to a paid campaign. A creator who features an unsolicited product carries a disclosable material connection and frequently discloses it the same way as a paid integration, so where the distinction affects a decision — rate benchmarking especially — confirm the terms directly rather than inferring them.
The result is a public-evidence classification
A high-confidence classification can support market intelligence, creator research, and campaign planning. It should not be represented as legal proof of a contract or as a complete record of every undisclosed commercial arrangement.
See sponsorship classification in market context
Bring a brand, creator, or category and see how qualified relationships become competitive intelligence.
Book a Demo