How we use AI, and where we stop.
We build synthetic presenters, AI-assisted articles, and generated product imagery. We also run a network of real human creators. This page explains which is which, how each is labelled, and the one line we don't cross.
Disclosure is the product, not the tax
We sell to security practitioners. They are, professionally, people who check things. An audience that discovers undisclosed synthetic content does not conclude that the content was clever — it concludes that the claims might be too.
So we disclose everywhere disclosure is available, whether or not the platform enforces it, and whether or not detection would catch us. That is cheaper than the alternative and it is the only posture consistent with selling security.
Synthetic presents.
Human testifies.
A synthetic presenter can state a sourced fact about a product. It cannot report an experience, give a testimonial, or claim to have used anything. Experience requires a person who actually had one.
Every other rule on this page follows from that sentence. When something is ambiguous, we resolve it by asking whether the content is presenting information or vouching for it.
What we do and don't do with AI
We use AI for
- ›Synthetic presenters — trained identities that present product and service information, always disclosed.
- ›Product and environment imagery — generated from real product photography as reference.
- ›Article drafting and structure — with human editing and human-verified facts.
- ›Podcast narration — from published written source material.
- ›Translation and dubbing — of existing approved content.
- ›Slideshows and carousels — assembled from approved claims.
We never use AI for
- ›Testimonials or reviews. A synthetic person has no experience to report.
- ›Faces resembling real, identifiable people. Ever.
- ›Cloned voices of real people without written consent.
- ›Unsourced factual or comparative claims. Every fact traces to a document.
- ›Presenting synthetic personas as independent reviewers or as staff.
- ›Assessment findings, incident details, or client outcomes. Those are reported by the humans who did the work.
Articles
AI assists with drafting and structure. A named human edits, verifies every factual claim against a primary source, and is accountable for what publishes. Articles carry a production note where AI was materially involved.
Images
Product imagery is generated using the manufacturer’s own photography as reference, so geometry and markings match the physical product. Where a rendered image could be mistaken for a product photograph on a listing, we use the manufacturer’s photograph instead. Generated imagery is for editorial, social and advertising use — not for the buy box.
Podcast and voice
Episodes narrated by synthetic voice are labelled as such in the episode description. We do not clone a real person’s voice without their written permission, and we do not present a synthetic voice as a named human host. Where a real person speaks, that is the default and it is not labelled, because it needs no label.
Human creators
Our creator network is people. Creators may use AI tools in their own production, provided a named human is accountable for the account and for the claims. A creator endorsement means a person used the product. An account with no accountable human behind it may still earn affiliate commission, but it is flagged and never presented to a partner brand as a human recommendation.
The tooling behind this — Remotion, Higgsbee, NanoBanana 2, Claude + MCP pipelines — is described on Creatives. The human side, including the creator-facing version of these rules, is the Creator Network.
Where synthetic content can go
Platform rules differ more than most people assume, and enforcement models differ even more. Some detect automatically, some rely on self-declaration, and at least one is reported to suppress reach on AI-generated organic content regardless of disclosure.
| Platform | Requirement | Enforcement | Our practice |
|---|---|---|---|
| TikTok | RequiredAIGC toggle at upload | Automated — reads C2PA content credentials, can auto-label, throttle or remove | Always toggled. Assume detection regardless. |
| YouTube | Requiredfor realistic altered or synthetic content | Creator self-disclosure at upload; enforced since early 2025 | Always disclosed on any presenter video. |
| MetaFacebook · Instagram | Requiredin Ads Manager for synthetic or AI-manipulated ad creative | Classifiers plus metadata partnerships; graded by how photorealistic the content is | Disclosure control enabled on all synthetic creative. Photorealistic synthetic humans trigger the most prominent labelling, so product-led formats are preferred. |
| Google Ads | Requiredwhere creative depicts real people synthetically | "AI Generated" badge applied to qualifying formats | Disclosed. We do not depict real people at all. |
| Requiredfor advertisers in Campaign Manager | No automated organic detection reported; reach suppression on AI content has been reported | Human-authored only. We do not post synthetic presenters here. | |
| X | Requiredfor ad creative | Lighter-touch labelling | Disclosed in creative. |
| CyberAdX networkowned properties | Our ruleDisclosure on every synthetic creative served | Our own advertising policy — enforced at campaign review before a creative goes live | Synthetic creative is labelled and logged against the campaign record. |
| Amazon | Verifyper listing-image standards | Listing image requirements govern product photography | Manufacturer photography only on listing images. Generated imagery is not used in the buy box. |
| eBay | Verify— | Listing image standards | Manufacturer photography only on listings. |
Two rules we apply beyond what platforms require.
LinkedIn is human-authored. It is where practitioners evaluate credibility, its enforcement model is the least predictable, and reach suppression on AI content has been reported there. The upside of posting a synthetic presenter on LinkedIn does not cover the downside.
Marketplace listing images are manufacturer photography. A generated image on a buy-box listing is a returns problem before it is a policy problem. Generated imagery earns its place in social, editorial and advertising.
Platform policies change frequently. This table was compiled 19 August 2026 and is reviewed quarterly. Where a row reads “verify”, we confirm current policy before publishing rather than assuming.
Beyond platform rules
Platform policy is the floor. Three regulatory vectors sit above it and they move faster than the platforms do.
EU AI Act
Transparency obligations for AI-generated content became applicable in 2026. Generative AI used in advertising sits in the limited-risk tier, where the core duty is transparency: AI-generated content shown to EU users must be disclosed. Our vendors are European and much of our audience is too, so we treat EU transparency as the baseline everywhere rather than maintaining two standards.
FTC Endorsement Guides
The FTC’s expanded guides and its enforcement work on AI claims address fake reviews, fabricated testimonials and undisclosed material connections. This is the specific reason our synthetic presenters never testify — a fabricated customer experience is the exact harm those rules exist to prevent, and disclosure does not cure it.
Other jurisdictions
Requirements for labelling AI-generated advertising and for rules on virtual endorsers are tightening in several markets. Where we publish into a jurisdiction we do not know, we ask before we publish.
What clients agree to
Every statement of work for presenter or content services carries these terms. They are not negotiable, and we would rather lose the engagement than vary them.
- 01The client discloses AI-generated content per each platform's requirements.
- 02Synthetic presenters are never presented as real customers, independent reviewers, or employees.
- 03The client supplies or approves every factual and comparative claim. We do not invent specifications.
- 04Comparative claims about competitors are attributed to that competitor's own documentation, or they do not run.
- 05No presenter resembles a real, identifiable person.
- 06The client indemnifies for claims it supplies.
We also say plainly which parts of a brief we think will not work. Products under roughly two centimetres do not composite convincingly into a presenter's hand; regulated claims need a source before they need a script. Being told this at scoping is cheaper than discovering it at review.
These terms are reproduced on every spec sheet the configurator produces, and they sit alongside the network-wide advertising policy.
If something looks wrong
If you see content of ours that appears undisclosed, misattributed, or factually wrong, tell us and we will correct it publicly rather than quietly. Corrections are logged with the date and what changed.
If you want to know whether a specific piece was AI-assisted, ask. We will tell you exactly which parts and which tools.
creatives@cyberadx.network