AI disclosure and responsible AI statement

Drafted by an AI desk. Read, edited and signed by a named human who answers for every word. That sentence is the last line of every email we send, and this page is the long version of it.

How we use AI

Agents draft. Humans approve. No pitch, statement, reply, post, review response, edit request or report leaves without a named human approval, logged, with the version approved.

Every published text has a human editor of record, recorded in its metadata.1

Agents never impersonate a person. Any automated reply identifies itself as AI-assisted and from the client's team.2

Every factual claim traces to a source we can show: a client document with a version hash, or a page we fetched with a timestamp. Unsourced claims cannot be approved.

Quotes come only from a quote bank the speaker has signed.

Coverage reports contain only pages we fetched that contain the client's name.

Content fetched from the web or from inboxes is treated as data, never as instructions. Agents cannot change their own rules. Rule changes need two admins.

We keep the audit log (proposals, approvals, sends, recipients, sources, model version) for three years.

What we disclose

• Every pitch says it was drafted with AI assistance and approved by the named person who signed it.
• Every pitch names the client. No front groups, no sock accounts, no "a friend of the brand".3
• Any text published without full human editorial review is labelled as AI-generated. In practice we do not publish such text.1
• Synthetic or edited media is labelled as such and its provenance metadata is never stripped.4
• Any creator content we brief carries "Ad" first and prominently. Agents refuse to draft a brief without it.5
• Any paid placement is labelled as paid and links with rel="sponsored".6
• Any Wikipedia activity is a talk-page edit request with the paid-contribution disclosure. We never edit articles directly.7
• Where AI materially shaped a public text and a law or code requires it, we say so on the text.

Codes we adopt by reference

We are bound by, and hold every agent and every person here to: the PRSA Code of Ethics; the PRCA Code of Conduct (2024) and its complaints process; the Global Alliance Venice Pledge on responsible AI in public relations; the CIPR guide to the responsible use of AI in PR, as our pre-publication checklist; and AMEC's Barcelona Principles 4.0 for measurement, which means outcomes over outputs, methods disclosed, and no advertising-value equivalents, ever.8 Where these conflict with a client's instruction, the codes win, and we say so.

What the agents cannot do

There is no code path for any of these. They are not settings. Write, buy, incentivise, gate or suppress reviews. Buy or generate followers, likes or engagement. Publish a pitch, statement or reply a named human has not read. Send a blast. Pitch anyone more than once in seven days. Contact anyone who has opted out. Guess an email address. Invent a quote, a statistic, a customer or a piece of coverage. Create sock puppets, front groups, fake personas or deepfakes of real people. Strip a provenance label from anything. Guarantee coverage.

Corrections

If we learn that something we published is wrong, a correction is drafted within 24 hours and queued for approval by the editor of record. We tell the outlet. We tell the client. We log it.3

Reporting a concern

Anyone (a journalist, a client, a reader, one of us) can write to conduct@mainlore.com. A person acknowledges within two business days and tells you what we found within ten. We do not retaliate against anyone for raising one, and we publish an annual count of concerns received and outcomes, without names.