Human-Made Is a Recommendation Rule, Not a Universal Label

An AI label can describe one part of a work’s production history. A recommendation rule answers a different question: whether a platform will distribute it beyond the creator’s existing audience. Snap has made that separation unusually visible in its recent Spotlight policy. Its choice to favor human-made creativity is a platform rule about reach, not a universal label for every image, video, or piece of writing made with AI.

That distinction matters because “AI-generated” can cover a range of acts, from a wholly synthetic video to an edit that removes background noise. A platform marker, a legal disclosure duty, and an eligibility decision may each describe something different. Treating them as interchangeable can make a recommendation policy sound like a certificate of truth or a general ban on synthetic media.

Three questions, not one

The first question is about the work: how was it made? A content label may identify a generated or manipulated image, audio clip, or video. But a label does not automatically describe every step in the production chain. It may not tell a viewer whether a person supplied the central idea, wrote the script, selected the shots, edited the result, or approved the final version.

The second question is about rules. Snap’s Community Guidelines apply across Snapchat, while its separate Content Guidelines for Recommendation Eligibility add stricter standards for material that could reach people beyond the creator’s friends or subscribers. The policy names Spotlight, Public Stories, and Snap Map as surfaces where those additional standards apply; Chat and Private Stories are not subject to that extra recommendation rule, though the baseline Community Guidelines still apply. Snap says its guidelines apply to content from individuals, partners, and organizations alike. Snap’s recommendation-eligibility overview

The third question is legal. The European Commission’s guidance on Article 50 of the EU AI Act describes separate duties for providers and deployers: providers have obligations to mark certain generated outputs in a machine-readable form, while deployers may need to disclose certain deepfakes or public-interest text made without human review or editorial control. The guidance also describes scope limits and exceptions. Those are legal duties for defined roles and circumstances, not a worldwide rule that every AI-assisted post must carry the same public label. Snap’s Community Guidelines likewise say that some regions require users to disclose or label AI-generated content, but the page does not identify each jurisdiction or map its requirements. The Commission’s Article 50 guidance Snap’s Community Guidelines

Snap’s rule has a surface

In a July 31 newsroom post, Snap said that wholly AI-generated videos would no longer be eligible for recommendation on Spotlight, while content enhanced or edited with Snapchat’s AI creative tools would remain eligible and include transparency indicators. This is Snap’s public description of its product decision, not an independent test of how often the rule is applied or how well the detection works. The company also acknowledged that no detection system is perfect. Snap’s Spotlight announcement

The more detailed Quality page in Snap’s recommendation rules describes the treatment in slightly different terms. It says the ranking algorithm rewards authentic, human-made content over wholly AI-generated content created outside Snapchat, even if that content has a transparency disclosure. It says AI-generated content created within Snapchat is eligible for recommendation because the company’s tools include transparency indicators. These statements make origin and degree of AI involvement relevant to eligibility, but they do not explain a precise boundary for “wholly AI-generated,” describe a detection method, or give the weight of human-made content in ranking. Snap’s Recommendation Eligibility: Quality page

There is a meaningful gap between being eligible and being recommended. Eligibility means the content can enter a distribution process under the stated rules; it does not promise impressions, placement, or audience growth. Conversely, a rule that excludes a kind of content from a public recommendation surface is not the same as removing that content from every area of Snapchat. Snap’s overview makes the surface distinction explicit: private communication follows baseline rules, while recommendation beyond chosen friends or subscribers brings additional standards.

For a creator, the immediate operational question is therefore not “Is AI allowed?” in the abstract. It is “What kind of material is this, and which surface’s rule governs its reach?” A synthetic clip may be allowed to exist under baseline rules, carry a disclosure, and still be ineligible for broad recommendation. Those states can coexist without proving that the clip is false, unauthorized, or prohibited everywhere.

A marker and a ranking gate do different jobs

A transparency indicator is information about provenance or platform tooling. A recommendation decision is a distribution choice. One can be present without changing the other: Snap’s detailed rule says that an off-platform, wholly AI-generated work can remain ineligible even when it includes a disclosure. The marker may help viewers understand a production fact, but the platform’s eligibility rule still applies.

The reverse also matters. A piece that is eligible for recommendation is not thereby certified as human-authored, factually accurate, authorized by everyone depicted, or compliant with every law. Snap’s public materials do not claim that recommendation eligibility verifies those things. They describe policy standards and enforcement choices for a company-operated service.

Nor does a platform’s use of an indicator make that indicator universal. Snap says its own AI tools include transparency indicators; its broader Community Guidelines place responsibility for regional disclosures on the person sharing synthetic media. The public policy pages do not establish that every off-platform editing tool inserts a comparable marker, that markers survive every export, or that a missing marker proves human-only production. Absence of a signal is not evidence of absence of AI use.

This is where law, company policy, and evidence should remain in separate columns. Law can require a defined provider or deployer to disclose certain generated or manipulated material in a particular context. A company can set a stricter distribution rule for one product surface. A visible indicator can provide a clue about a workflow. None of those facts, by itself, resolves questions about a depicted event, a person’s consent, rights to source material, or the identity and intent of the poster.

Four questions for reading the boundary

When a post is described as “human-made,” “AI-generated,” or “AI-assisted,” ask what claim is actually being made:

• Which surface? Is the claim about private sharing, public availability, or recommendation to people who do not follow the creator? Snap’s additional eligibility rules do not apply to every feature in the same way.
• What work is described? Is the material wholly generated, manipulated to depict something as authentic, or edited with an assistive tool? A broad label may collapse materially different production steps.
• Which rule is doing the work? Is this a legal disclosure, a platform’s community rule, an eligibility condition, or a marker supplied by a particular tool? Each has a different purpose and scope.
• What does the evidence support? A policy page documents what a platform says its rule is. It does not independently establish that every decision is correct, that the system detects every case, or that a particular post was made in a specific way.

That reading preserves a useful tension: platforms can make choices about what they recommend without defining authenticity for the entire internet. A “human-made” preference may shape a distribution system, but a recommendation decision is not a universal label. Readers still need provenance, context, and evidence for the separate claims that matter to them.

Sources and limitations

• Snap, Content Guidelines for Recommendation Eligibility — checked October 21, 2026; supports which Snapchat surfaces receive additional recommendation standards and the distinction from baseline Community Guidelines. Limitation: this is Snap’s public policy description; it does not disclose every implementation detail, individual decision, or ranking weight.
• Snap, Recommendation Eligibility: Quality — checked October 21, 2026; supports Snap’s stated distinction between wholly AI-generated content created outside Snapchat and content created with Snapchat tools. Limitation: the page does not define “wholly AI-generated,” provide detection accuracy, or establish how every borderline edit is classified.
• Snap, “Rewarding Authentic Creativity on Spotlight” — checked October 21, 2026; supports the July 31, 2026 Spotlight announcement about wholly AI-generated videos and AI-enhanced or edited content. Limitation: company statement, not independent measurement of enforcement or recommendation outcomes.
• Snap, Community Guidelines — checked October 21, 2026; supports the baseline rules applying to content across Snapchat and Snap’s statement about regional AI disclosure requirements. Limitation: the page does not identify all jurisdictions, statutory elements, or effective dates.
• European Commission, Transparency obligations under Article 50 of the AI Act — checked October 21, 2026; supports the distinction between provider marking and certain deployer disclosures under EU rules. Limitation: Commission guidance is not a worldwide rulebook or advice for a particular case; roles, exceptions, and transitional details matter and should be checked against current legal text.