AI Transparency & Prediction Disclaimer

This notice applies to Coul's viral-potential scores, component scores, explanations, recommendations, AI-assisted edits, generated content ideas, and related automation. It supplements the Terms of Service, Privacy Policy, and User Content & AI Output Policy.

Coul's prediction system is separate from Instagram, TikTok, X, YouTube, and every other social platform. Those platforms do not issue, endorse, or receive a duty to follow Coul's score. A platform's own ranking, moderation, advertising, and recommendation systems operate independently and can change without notice.

The score is Coul's model-generated estimate of how selected content characteristics compare with patterns available to the configured system. Unless a result expressly states a calibrated probability, prediction target, baseline, observation period, and validation date, a 0–100 score is a relative decision-support index—not a percentage chance that a post will go viral.

There is no universal legal, platform, or industry definition of “viral.” Before a production score is shown, Coul must identify the outcome it is estimating, the relevant platform or content type, the measurement horizon, and the comparison baseline. A score is not a view forecast, percentile, quality grade, platform ranking decision, or factual statement about future performance.

Scores from different model versions, platforms, niches, languages, account contexts, or dates may not be directly comparable. Re-running the same content can produce a different result when the model, available context, or data changes.

Depending on the feature, supported format, permissions, and information available, Coul may analyze some—not necessarily all—of the following. These are feature-dependent potential inputs, not a claim that every item is used for a particular score. The current scoring worker receives a source reference—such as a URL or media identifier—and source metadata; audience, niche, account history, feedback, workspace memory, and trend context must not be described as used unless the production request and report actually identify them. Each result should identify the categories used and mark important information that was unavailable.

• Opening hook, narrative or visual structure, pacing, cut timing, duration, format, composition, movement, and information density.

• Speech, transcript, caption, on-screen text, language, topic, clarity, sentiment, calls to action, and accessibility-related signals.

• Audio characteristics, silence, music or beat timing, and audio-visual alignment. The score is not designed or permitted to identify a person from their voice.

• Audience, niche, stated goals, intended platform, publishing context, and settings you provide.

• Your authorized account history, prior content and performance data, feedback, and relevant workspace memory when available and permitted.

• Lawfully available aggregate or deidentified trend and performance patterns. Competitor posts can inform inspiration and contextual patterns but are not eligible for content-level scoring or editing merely because they are public.

A report may include an overall 0–100 score and separate indicators for quality, hook strength, pacing, retention risk, transcript or beat analysis, plus recommendations and explanations. These fields answer different questions and should not be combined as if each were an independent probability.

Explanations identify patterns or features the system associated with its result. They are model rationales—not proof that a feature caused past performance or that changing it will cause future success. A recommendation can raise an internal score without producing the same real-world result.

Estimated views, engagement, lift, conversions, or revenue must not be presented as a production prediction unless Coul separately validates that outcome, shows a range rather than false precision, and identifies the baseline, horizon, method, uncertainty, and material assumptions.

AI predictions contain uncertainty. A production result must show the model or scoring version and evaluation date, the supported platform and content type, relevant data freshness, and a confidence or evidence-status indicator. When context is too weak, Coul should say “insufficient evidence” instead of displaying unjustified precision.

ModelMetadata that is optional, hidden, or free-form is not an adequate user disclosure. Coul must bind required provenance fields to every production report and preserve the version used so later feedback, disputes, and re-runs can be understood.

Coul should maintain evaluation records appropriate to each supported platform, language, niche, and content type. All objective mechanism, validation, calibration, causal, and performance claims require documented and current substantiation or clear qualification before publication. This includes qualitative claims such as analyzing every second, learning continuously, or changing an outcome, as well as numerical claims about accuracy, average lift, videos analyzed, views, or time saved. A general disclaimer does not make an unsupported objective claim accurate.

Coul can miss context, misunderstand speech or visuals, rely on incomplete or stale information, overvalue a correlation, or produce a score or edit that is inconsistent, biased, unsafe, or simply wrong. AI-generated text can also invent facts, sources, quotations, rights, or platform rules.

• Platform algorithm, policy, format, API, and moderation changes; outages; regional differences; and unavailable or deleted metrics.

• Posting time, trend decay, breaking events, novelty, geography, audience fatigue, creator reputation, account size or history, paid promotion, and authentic versus fake engagement.

• Sparse history for a new creator or niche; unsupported or mixed languages; inaccessible private posts; corrupted uploads; no or muted audio; unusual formats; low-quality transcripts; and incomplete captions.

• Copyright, music, sponsorship, advertising, community-guideline, and legal restrictions that performance data does not reveal.

Performance can be uneven across languages, dialects, regions, cultures, niches, accessibility styles, editing conventions, account sizes, and creators with limited history. Historical popularity can reflect unequal access, discrimination, platform enforcement, or manipulation; correlation with prior engagement is not a value judgment about a person or community.

Do not remove captions, audio description, readable text, sufficient timing, or another accessibility feature merely to increase a score. Do not use a lower result to stereotype a language, culture, disability, appearance, or audience. Media involving minors or vulnerable people requires heightened human review and all appropriate rights and safeguards.

Coul is intended for adults who are at least 18 and the age of legal majority where they live. Regional notices, choices, or AI disclosures apply where required, and a feature may be limited where Coul cannot meet a local requirement.

Submit content for viral scoring or AI editing only when it is yours or you have the creator's or rights holder's permission to analyze and edit it. A public URL does not establish ownership, permission, or public-domain status. Competitor videos displayed in Coul are for lawful inspiration, comparison, and original ideation—not content-level virality analysis, copying, or automatic editing.

Before launch, every external-URL analysis path must require and retain an ownership or authorization attestation, accept only actually supported sources, and block known competitor-library items. An external URL lacking that control must not enter the scoring worker. The present policy rule does not substitute for a technical gate.

Use patterns, themes, and formats as inspiration while creating original expression. Do not reupload another person's media, imitate protected expression too closely, impersonate a creator, remove attribution or rights information, or suggest sponsorship or affiliation that does not exist.

Where an assisted editing feature is actually available, Coul may recommend or apply cuts, timing changes, captions, text, audio, framing, ordering, or other transformations to authorized media. A recommendation is not evidence that Coul has cleared the edited content for accuracy, rights, safety, accessibility, advertising, or platform compliance.

Compare the original and proposed version before accepting an edit. Check factual meaning, chronology, context, quotations, subtitles, translations, sponsorship and affiliate disclosures, legal disclaimers, calls to action, music and footage rights, voices and likenesses, brand requirements, and accessibility. Preserve an original or reversible version whenever the product supports it.

Coul must not market automatic cutting or editing as a live production feature until the implemented pipeline, preview, reversibility, version history, and approval controls have been verified. A simulated “apply” interaction must be clearly labeled as a demo.

Analysis, generation or editing, scheduling, and publishing are separate actions. Approval of an edit is not approval to publish it. Coul's current publishing safeguard binds human approval to a fingerprint of the title, caption, script, hashtags, selected platforms, format, and media; changing one of those values makes that approval stale. The destination account, scheduled time, and time zone are not currently part of that fingerprint and must be separately revalidated and reviewed before publishing after a change.

Daily AI suggestions do not publish by themselves. Automatic or scheduled publishing may occur only after you explicitly enable the applicable automation, select the destination and conditions, and approve the publishing snapshot required by the product. Review queued work and connected accounts regularly, especially after a team, role, token, or campaign change.

Even after approval, a platform can reject, delay, modify, remove, limit, or misdeliver a post. Check the final platform result and promptly pause automation if the output or destination is wrong.

When Coul generates or materially manipulates image, audio, video, or text, users must preserve and apply provenance, labels, watermarks, metadata, or platform disclosures supplied by Coul and add any further disclosure required by law, platform rules, advertising rules, or context. Do not remove a required machine-readable or visible marker.

Coul must support technically detectable provenance, machine-readable marking, user disclosure tools, or visible disclosure only where the applicable rule assigns that duty to Coul in its actual role as a model provider, downstream integrator, deployer, hosting platform, or another covered actor. Requirements differ by role, feature, audience, scale, and jurisdiction. Assistive standard editing that does not substantially alter input or meaning may be treated differently, but users must still disclose deception-sensitive changes, synthetic people or events, and deepfakes when required or reasonably necessary to avoid misleading an audience.

Do not generate or publish deceptive synthetic media, impersonation, non-consensual intimate imagery, fabricated evidence, false endorsements, or content that misrepresents a real person's actions or words. The Acceptable Use Policy applies even if a platform would allow the post.

Coul's viral score and creative recommendations are not designed or validated to make a legal or similarly significant decision about a person. Do not use them to decide employment, credit, housing, insurance, education, immigration, benefits, medical treatment, financial eligibility, political access, law enforcement, or another high-impact entitlement or opportunity.

Coul does not provide medical, legal, financial, investment, political-campaign, or other regulated professional advice. A score does not establish truth, safety, legality, authenticity, audience suitability, fair use, non-infringement, or compliance. Obtain qualified review when the content or decision carries material risk.

Never increase sensationalism, fear, harassment, deceptive engagement, unsafe acts, infringement, or unlawful targeting merely because a model predicts higher engagement. The Acceptable Use Policy controls prohibited content and conduct.

Coul does not guarantee any number of views, impressions, reach, ranking position, watch time, retention, likes, comments, shares, saves, followers, leads, sales, conversions, sponsorships, advertising approval, platform distribution, or revenue. Past performance, competitor performance, examples, testimonials, and a high internal score do not guarantee a similar result.

Your actual outcome depends on content, audience, account history, timing, platform behavior, competition, budget, rights, compliance, world events, and many other factors outside Coul's control. Do not spend money, make contractual promises, or forecast income based only on a score or recommendation.

This page supplements—not replaces—clear disclosure beside the score, edit, approval, and automation control. A result screen should show the short no-guarantee notice, target, supported platform and format, version, date, evidence or confidence status, material missing context, and a feedback route without requiring the user to find this document.

As of this notice's effective date, Coul's current prototype includes fixed “AI score” results selected by input mode and randomized engagement or “Viral Score” values in a URL-processing surface. Those values are illustrative demos—not production analyses or predictions. They must be visibly labeled as such immediately and gated, replaced with validated production output, or removed before launch.

Any fixed, randomized, staged, placeholder, or sample score must be unmistakably labeled “Illustrative demo—not an analysis or prediction” and visually separated from production results. It must never be used as if it came from uploaded content. Unsupported source types must not be presented as analyzed.

• Remove or relabel mock and random results before production use.

• Enforce the ownership or authorization gate on every file and URL path.

• Require version, target, horizon, platform, freshness, and uncertainty metadata for every production report.

• Substantiate, qualify, or remove objective mechanism, validation, calibration, causal, performance, and research claims before publication.

• Verify assisted-edit behavior, preview, reversibility, approvals, and provenance marking before marketing those functions as available.

Use the result feedback control, when available, to report a wrong transcript, unsupported language, missing context, harmful recommendation, rights concern, or unexpected score. Include the report identifier and model version but do not email confidential media or credentials. Where available, correct the context and run a new analysis rather than treating a past score as current.

Coul may update models, inputs, supported platforms, thresholds, evaluation methods, and this notice. Material changes to the meaning of a score or use of personal data require appropriate notice and, where law requires it, a choice before the change applies. Historical reports should retain enough version information to explain what generated them.

For data use, deletion, or model-training questions, see the Privacy Policy and Data Rights page. For copyright, ownership, or output questions, see the User Content & AI Output Policy and Copyright/DMCA Policy.