What AI can (and can’t) do for your brand.

We are not anti-AI. We use it constantly — for first drafts, image variations, research, copy iteration, even early-stage video work. But the gap between “AI can generate this” and “AI can ship this” is wider than most people admit. Here is the pattern we watch clients learn the hard way, project after project — and the honest mapping we use to spare them most of it.

Most articles about AI in branding are written by people who do not use it daily. They are either evangelists (everything is fine, AI will replace your team next quarter) or alarmists (everything is collapsing, AI cannot do anything). Neither matches our experience.

Here is the honest mapping. The places AI saves us hours every week, and the places it costs us hours we did not expect.

What AI is genuinely great at

This is the part the alarmists miss. AI is, right now, in 2026, dramatically better at certain creative tasks than any tool we have ever had.

First drafts of long-form copy. Press releases, FAQ pages, internal documentation, product descriptions. Claude and ChatGPT can produce a draft in 30 seconds that would have taken a writer 90 minutes. The draft is rarely shippable. It is almost always a useful starting point.

Asset variations at scale. Generating 20 versions of an ad headline, 15 variations of a banner image, 8 different color palette explorations for a brand exercise. The volume that used to be expensive is now cheap.

Research and synthesis. “Read these 12 PDFs of competitor brand strategies and tell me what they have in common.” AI is excellent at this. A human analyst can validate the output in 20 minutes instead of doing 8 hours of reading.

Transcription and translation. Client calls, meeting notes, multilingual subtitles. We use AI for the first pass and a human reviewer for the final. The error rate is low. The time savings is large.

Code scaffolding and content audits. Crawling a site, finding broken links, generating sitemaps, writing initial CSS for a layout. AI does the boilerplate. Humans do the judgment.

AI gets us to 80% on the boring 80% of the work. It cannot get us to 100% on the interesting 20%.

The video editing trap

This is the part I want to spend the most time on, because it is the most overpromised and the least honestly discussed corner of AI right now.

AI video tools — Sora, Runway, Veo — are stunning in their demos. In production they are an exercise in frustration management.

Here is what actually happens when we use them on a real project: the prompt produces a clip that is 85% great. There is a hand with the wrong number of fingers. The lighting drifts halfway through. A background element morphs into a different background element for one frame. The brand color is technically right but the saturation is slightly off. The motion is uncanny.

Each of these is a small problem. Cumulatively, they are why every “AI-generated” video ad you see has been touched extensively by a human in post. We routinely run 30-50 prompts to get the closest-to-usable clip, and then the clip gets pulled into Blender or Premiere or DaVinci Resolve for color correction, frame interpolation, mask work, and selective re-rendering of the parts the AI got wrong.

The math is uncomfortable. A 15-second ad spot that took 4 days to shoot the old way might take 2 days of prompt iteration plus 1.5 days of human cleanup. The savings is real but smaller than the demos suggest. And the cleanup work is more technical than the original shoot, which means the team you need is different (and arguably more expensive).

What AI is bad at (and probably will be for a while)

These are not edge cases. They are everyday tasks where AI consistently disappoints.

True brand voice. Brand voice lives in a thousand micro-decisions — sentence rhythm, word choice, when to break a rule. AI imitates the surface and misses the texture. We have run experiments. Even a 50,000-word style guide does not get a model to write reliably in a specific brand voice. A senior writer who has lived with the brand for two years does it without trying.

Sustained consistency across hundreds of pieces. AI generates each piece independently. Even with the same prompt, you get drift across 100 outputs. Humans, with the same brief, do not drift. HBR has covered this in the context of automated content production: scale is exactly the place where AI’s inconsistency compounds.

Judgment calls. “Is this funny or is it offensive?” “Does this client really mean what they said in the brief?” “Should we push back on this color choice?” These are not generative tasks. They are taste and politics. AI does not have taste. It has averages.

Original photography of real people. If your brand depends on real humans — your team, your customers, your patients — AI cannot give you them. It can give you fakes. The fakes are getting better. They will probably never be good enough to fool the customer who is in the room.

The honest workflow we use

For most projects, the actual breakdown looks like this:

1. Strategy and creative direction: 100% human. AI does not pitch ideas. It refines them.
2. First-draft copy: AI generates, human edits. Time saved: ~40%.
3. Asset variations: AI generates 20 options. Human picks 2-3 to refine.
4. Photography of real people: 100% human. No exceptions.
5. Video production: Real shoot + post production for most pieces. AI for explainer animations and pre-viz. Hybrid for stock B-roll.
6. Final QA: 100% human. Always. We do not ship work that hasn’t been touched by a senior creative.

This is not the workflow the AI companies advertise. It is the workflow that actually produces shippable work for real clients without embarrassing anyone six months later.

What this is really about

The discourse around AI in 2026 is unhelpfully binary. Either AI is going to replace creative agencies, or AI is a passing fad. Neither is true. AI is a tool. It is better than every previous tool at certain tasks and worse than every previous tool at others. The agencies that win are the ones that figure out the actual mapping — and are honest about it.

If your current agency tells you they “don’t use AI” they are either lying or they are inefficient. If they tell you they use AI “for everything” they are about to ship you something that does not work. The honest answer is in the middle, and it gets less interesting every quarter as the tools improve and the use cases solidify.

We use AI. We also use Blender. Both, on the same projects, often on the same day. That is the work in 2026.