Getting permission to show AI work publicly
Clients happy to run generated work often refuse to be named for it. Negotiate case-study permission at signature, not after delivery.
Notice how many AI production case studies describe an unnamed client. A mid-size e-commerce brand. A DTC company in the wellness category. A regional retailer. The numbers are often specific and the client never is, and that pattern is not modesty. It is what a portfolio looks like after the permission conversation happened too late.
The sequence is consistent. The client is delighted with the work. They run it. It performs. You ask whether you can write it up, and the answer comes back as a soft, apologetic no from someone who is not the person you have been working with.
Why they refuse, and why it is not about your work
The refusal almost never comes from the marketing lead you built the relationship with. It comes from communications, legal, or a brand director who has one specific fear and no appetite for testing it.
The reference case in their head is usually Coca-Cola's holiday campaign, which compressed a production timeline of roughly a year down to about a month and drew significant public criticism for the method at the same time. Executives read that as: the work was fine, the disclosure was the problem, and the brand absorbed the reaction. Nobody wants to be the second example.
The underlying sentiment data supports the caution. IAB and Sonata Insights research published in January 2026 put advertising executives 37 points apart from Gen Z and millennial consumers on AI-generated ads, widened from 32 points in 2024, with 60% of US ad professionals citing accuracy and transparency as a top barrier to adoption. A brand director looking at a widening gap is not going to volunteer their logo for a post about generative production, no matter how good the work was.
Two things follow. First, this is not a judgement of your craft, so do not negotiate as though it is. Second: the objection is almost always to being named in a story about AI, not to being named at all, and not to the work being shown. Those are three separable permissions, and separating them is the technique.
Some brands do go public. Klarna has stated publicly that AI accounts for around $10 million a year in savings across image production and external marketing suppliers. But large organisations have communications capacity to manage the reaction. Your client probably does not, and the tier system below is how you get most of the value anyway.
Four tiers, not one question
Asking "can I use this as a case study?" invites one answer and it is usually no, because the person answering has to imagine the maximum version. Offer a ladder instead.
| Tier | What you may publish | What the client is agreeing to |
|---|---|---|
| Named | Client name, marks, assets, a quote, results | Public association, reviewed before publication |
| Anonymised | Assets, category, company size, results, no name | Their work shown without attribution |
| Method-only | Process, workflow, outcome shape, no assets, no name | Nothing identifiable leaves the building |
| Private reference | Assets and name in pitch materials under NDA only | No public exposure at all |
Almost every client says yes to at least tier three, and tier three is worth far more than nothing. Anonymised assets carry a showreel. Method-only carries a services page and a pitch narrative. Private reference carries the pitch meeting, which is where deals actually close.
Ask for the ladder rather than the top rung and you convert a binary into a selection, which is a much easier decision for someone whose default is caution.
The clause, at signature
Put it in the statement of work with the tier as a selected field, not a negotiation. Clients tick boxes readily and open discussions reluctantly.
Publicity. Supplier may reference the engagement and display the Deliverables in its portfolio, showreel, website and pitch materials, at the tier selected in this Statement of Work, from [30] days after first public release of the Deliverables by Client.
Review. Where the Named tier is selected, Client approves use of its name and marks for this purpose. Case study text and any results figures are provided to Client at least [5] business days before publication. Client may correct factual errors and may withdraw any specific figure.
Confidential material. Supplier will not disclose unreleased products, pricing, internal strategy, or performance figures that are not already public or expressly approved.
Withdrawal. Client may withdraw permission for a specific item on written notice. Supplier will remove it from active channels within [10] business days. Withdrawal does not apply retrospectively to printed or already-distributed materials.
Downgrade. If Named permission is withdrawn, permission continues automatically at the Anonymised tier unless Client states otherwise in writing.
The downgrade paragraph is the one that earns its place. Without it, a withdrawal is total and you lose the asset from your reel entirely. With it, a change of heart at the communications layer costs you the logo and leaves you the work, which is the larger part of the value.
The delay before publication matters too. Thirty days after the client's own release means you are never the first party to make the work public, which removes the single most common reason a marketing lead has to say no.
How to ask so it does not become a project
The ask belongs in the same conversation as scope and rates, treated as ordinary supplier terms rather than a favour.
"One admin item in the SOW: portfolio permission. There are four levels and you pick which one, so nothing is open-ended. Most clients go with the second, which means I can show the work without naming you. If you're comfortable being named I'd love that, and you'd see anything before it goes out. Either way, nothing publishes until thirty days after you've run it, and you can pull an item at any time."
Three things that script does. It names the safe default before the ambitious ask, so the ambitious ask is a step up rather than a request from zero. It puts the review right and the withdrawal right in the same breath as the permission, which removes the two objections that would otherwise come back a week later from someone else. And it frames the whole thing as an admin item, which is how it should read, because a permission that only gets granted when it feels like a big favour is a permission you will not get.
If the client selects Named and legal later downgrades it, that is a normal outcome and the clause already handles it. What you have avoided is the version where you ask after delivery, get a no, and have nothing.
What a hard no still leaves you
An anonymised case study is not a consolation prize if you build it properly. The reason so much AI production reporting features unnamed clients is that anonymised work still sells, provided the specifics are real.
What carries it: the category and rough size, the constraint you were working against, the production approach, the timeline, and results the client has approved as figures even if they will not approve their name against them. What kills it: vagueness. "A leading brand saw significant uplift" is worth nothing. "A twelve-person DTC brand in the supplements category, six weeks from brief to live, twenty-two variants tested" is worth a meeting, and none of it identifies anyone.
Two other routes worth having in place.
Build part of your reel from work you own outright. Spec pieces carry no permission problem and let you demonstrate the capability you want to sell rather than whatever the last client happened to need. Pitching brands with spec ads and spec video outreach for local clients cover producing those deliberately, and building a portfolio reel covers the assembly.
And if you work through an agency rather than direct, check the pass-through terms first. White-label arrangements frequently prohibit attribution entirely, and finding that out after building a reel around the work is avoidable.
FAQ
Should I mention AI in the case study at all?
If the client is named, follow whatever they agreed and keep the emphasis on the outcome and the constraint rather than the toolchain. If the client is anonymised, describe the method freely, because that is the part with no permission attached and the part a prospective client is actually evaluating. Either way, the commercial conversation about your production method belongs in the sales process, not discovered from a case study; what to tell clients about using AI covers where that conversation sits.
Can I show work the client never published?
Only with explicit permission, and expect no more often than yes. Unreleased concepts are the category clients guard hardest, and unlike finished work there is no public version to point at. If you want unpublished concepts in the reel, ask for them specifically rather than assuming the general publicity clause reaches them.
What results figures can I use?
Only figures the client has approved in writing, and never anything derived from their analytics that they have not seen written up. Where a client will not release numbers, describe the shape instead: variants tested, time from brief to live, revision rounds. Those are yours to report and they are often more persuasive to a buyer than a conversion percentage they cannot verify. AI video ROI case studies is a useful calibration for which figures are worth chasing.
Is a testimonial easier to get than a case study?
Usually, yes, and it is worth asking separately at the point of highest goodwill, which is the week after a successful delivery rather than at the end of the engagement. A named quote about working with you carries most of the credibility of a named case study and asks the client to approve one sentence instead of a page. Turning written case studies into video covers what to do with them once you have them.