Guardrails
We don't wag the dog.
“Wagging the dog” means inventing a distraction to bury a story. It's a great name for a film, and a terrible way to run a reputation. The only dog in Conrad's office is the golden retriever asleep on his desk, and nobody wags him. Conrad's name is a joke. These rules are not.
What Conrad will never do
No tricks. Not even small ones.
Each of these is enforced in software, before a draft ever reaches a person for approval, and written into every client agreement.
- No fake people.No sock puppets, made-up personas or accounts that pretend to be someone they are not.
- No bots or fake crowds.No automated amplification, bought followers or coordinated networks posing as ordinary people.
- No fake reviews.We never write, buy, gate or bury reviews, in line with the FTC's 2024 rule on fake reviews and testimonials.
- No fabrication.No invented stories, quotes, evidence, studies or experts. No deepfakes, ever.
- No astroturf.No undisclosed paid endorsements or “grassroots” campaigns that are secretly ours, per the FTC Endorsement Guides.
- No distractions.We answer the story. We don't manufacture a different one to drown it out.
- No attacking critics.No doxxing, no smears, no threats meant to silence truthful speech.
- No hiding who we work for.When we pitch a reporter, we say who our client is.
What Conrad always does
The machine drafts. A person decides.
- A named human approves every post, pitch and reply. Nothing is published automatically.
- Every claim has a source. Drafts are checked against the facts you supply; anything unsupported is flagged, not published.
- Disclosure where it is due. AI-assisted content is labelled where the law or a platform's rules call for it.
- Corrections over cover-ups. If you got it wrong, Conrad will help you say so, clearly and early. It works better anyway.
- Everything on the record. Every alert, draft, edit, approval and refusal is logged, and the log is yours.
- Your data stays yours. Processed on private hardware, never sold, never used to train models for anyone else.
How it is enforced
A rulebook that can't be sweet-talked.
AI models can be persuaded. Rules can't. Every draft and every action Conrad proposes passes through Demi, a deterministic rules engine that sits above the AI and checks it against the guardrails above. If a request breaks a rule, it is refused and the refusal is logged, however it was worded and whoever asked.
Conrad will tell you no. If you ask for something on the list above, he will say so, explain why, and offer the honest version that usually works better. If you need someone to do it anyway, we are the wrong firm, and we would rather tell you on day one.
Want the honest version, fast?
Tell us what is happening. If we can help within these rules, we will say so today.