Before Your Team Adopts Generative AI, Write These Policies

It is already in the building. That is the premise here, and three weeks of conversations since the release confirm it everywhere. Writers are experimenting with little fanfare, some disclosed, most not. An executive has asked someone in your organization about content cost savings within the last fourteen days. The vendors are re-labeling their products by the hour.

The reference points are stacking up daily. Stack Overflow banned generated answers this month because moderators could not keep up with plausible-but-wrong submissions. That is the hallucination finding from our first notes arriving as a governance event, on schedule.

The policy vacuum every organization currently has is not neutral. It is filling itself, one unexamined decision at a time. In single writers’ habits. In unreviewed drafts entering pipelines. In precedents set by whoever adopted first. The December assignment for content leadership, our chair addressing its counterparts, is to choose positions before scale chooses them for you. The five decisions that need writing down follow. The house positions are disclosed where we have them and flagged where honest doubt remains, because performing certainty about December 2022 would be dishonest.

The five decisions that need writing down

One: disclosure. Will readers know when text is machine-drafted, and what does your answer say about your brand? The spectrum is real (from full public labeling, through internal-only disclosure, to silence) and the norms are unformed. The deciding consideration we keep landing on is the trust frame. Your institution’s content is believed because of who stands behind it. Disclosure policies are trust policies. The organizations whose relationship with their audiences depends on being real, which is this entire readership, should err toward candor and decide now. A disclosure policy adopted after an embarrassment is an apology, not a policy.

Two: quality gates. The standard holds regardless of who or what typed. August’s helpful-content classifier does not care who wrote the words. It grades whether content serves people, and the sitewide blast radius now prices unreviewed generated content as a domain-level liability. The gate that was always right (editorial review, fact checking, the expert pass) becomes mandatory at whatever volume generation enables. That means volume is gated by review capacity, not generation capacity. Any plan that scales output without scaling checking has misread the economics in the way the rankings will eventually correct.

Three: real expertise, really involved. The E-A-T logic meets fluent text without authors, and the resolution is the one the credibility work has been building toward for years. Your credentialed voices just became more valuable, not less. Machine-fluent prose is now commodity. Verified expertise is not. The policy question is how your named experts relate to drafted text. Review and adopt, with real involvement and real accountability. Versus byline theater, names attached to unread output. That will be this era’s version of the content-farm tell, spotted by audiences and, in time, by classifiers.

The remaining two, and the boundary that organizes everything

• Four: workflow placement, the central decision. Drafting help: the model as junior partner. First drafts, summaries, reshaping, ideas. Always under human judgment before anything moves forward. Versus the publication pipeline, generated text flowing toward readers with little or no review. The two carry profoundly different risk, and the boundary between them is the policy. Our house position, disclosed: drafting assistance is available to our teams now, under the review standards that were already mandatory. Nothing generated enters a publication path unreviewed, full stop. The boundary is written, named, and audited, because the drift from the first mode to the second is exactly the slope organizations slide down one deadline at a time.

• Five: legal, flagged honestly. The training-data and copyright questions are truly unsettled. The provenance of the corpus, the status of the output, the litigation already forming. Secrets have a new surface. What your staff paste into a third-party prompt box is a data event, and the acceptable-use part of the policy needs writing this quarter. Counsel’s discomfort is appropriate rather than obstructive. The correct approach is bounded use inside named constraints while the frameworks form, reviewed on a calendar.

The stance, and the method behind it

The position we are taking with clients, stated plainly because the season’s discourse offers only the poles. These tools are entering content work regardless of anyone’s enthusiasm or dismay. The capability is real, the pressure is economic, and the adoption is already underway in your building and ours. That converts the leadership question from whether to under what governance. And rules chosen on purpose beat rules discovered in a rankings drop, a credit embarrassment, or a leak, which are the three visible ways the vacuum resolves itself unattended.

The method is the standard one for arriving technologies, and readers will recognize every instrument. The sanctioned lane built before the sprawl, the no-code playbook applied to prompts. The boundary decided before the drift, the workflow placement above. The review capacity funded as the actual constraint. And the policy written to be revised. Dated, owned, calendared for quarterly review as the landscape moves, which it will.

None of this requires resolving the large questions our first notes left open. All of it is executable this quarter, by the content leadership this post addresses, with the documents it outlines. The organizations that do it will spend next year adopting on their own terms while the vacuum-governed spend it explaining incidents. Deliberate beats default, written beats assumed, and the meeting costs an afternoon.

Closing out 2022

The year-end paragraph, in the tradition, for a year that ended by changing the subject. 2022 ran the full range of our method. Platform milestones assessed, sunsets calendared, architecture fights refereed, drills run calmly. Then November arrived and handed every topic a new variable. The content economics the helpful-content coverage pre-read. The search architecture we have circled for years. The expertise moat a decade of credibility work built. All now live questions under a general capability whose first honest notes we wrote three weeks ago.

The policies sequenced here are the bridge between those notes and the working year ahead. Modest, written, revisable tools for working on purpose inside the unknown, which has been our entire product since 2006.

Write them this quarter. Review them quarterly. And enter the new year the way we intend to. Curious, governed, and in writing. The boundary named, and the credentialed voices standing visibly behind everything that carries the institution’s name. See you in January. Bring your policies.