Last month we wrote about AI writing every senior living website into the same voice. The same thing is now happening to the graphics, and it is moving faster, because a poster takes thirty seconds to generate and nobody has to read it before posting.
Scroll a local feed and watch it happen. The bakery two blocks over posts a warm, glowing autumn promo with soft bokeh and a script font. The handyman posts the same composition in blue. And then your competitor down the road posts an assisted living open house graphic built from the same template, the same lighting, the same vaguely European sans-serif, the same slightly-too-perfect stock-looking faces. Three completely unrelated businesses, one look.
The problem is not quality. It is attribution.
These graphics often look good. That is what makes the situation confusing for people inside the company. Someone on the team produces something genuinely striking in a few minutes, everyone agrees it looks better than what they were making before, and it goes out.
But a graphic’s first job is to be recognizably yours before anyone reads a word of it. If a family scrolls past your post and a competitor’s post in the same minute and cannot tell which was which, the design did not work, no matter how good it looked in isolation.
What actually goes wrong
It is almost never one dramatic failure. It is an accumulation of small drifts that nobody owns:
The fonts change every time. The model picks whatever looks good in that composition. Over a month you have used nine typefaces, none of them the two in your brand standards.
The colors are close but not right. A warmer red here, a navy that drifts toward teal there. Individually invisible. Side by side in a feed, it reads as a brand that does not know its own colors.
The logo is not your logo. This is the one that should stop a post from going out. Image models cannot reproduce a wordmark, so they generate something logo-shaped: your initials in a font you do not own, or a swoosh that was never yours. It looks intentional to someone who has never seen the real mark, which is exactly the problem, because that describes every prospective family.
The language does not sound like you. The headline on the graphic came from the same generator as the image, so it uses the same borrowed vocabulary as everyone else’s.
The place in the picture is not your community. A generated building with a generated courtyard sets an expectation your actual front entrance then has to meet on tour day.
Why it spreads inside a company
This is not a discipline problem, and scolding the team will not fix it. The tools removed the last friction that used to force a check. Before, a graphic went through someone who owned the brand, because making one required software and skill. Now the activities director can produce a beautiful poster on her phone between residents, and there is no natural moment where anyone asks whether it is on-brand.
The answer is not to take the tools away. It is to give people a lane narrow enough that the fast path and the on-brand path are the same path.
What we set up for clients
A small kit, not a manual. The actual logo files in the formats people will need, including one that works on a dark photo. The two fonts, installed, with the sizes already decided. The exact hex values. Three or four templates for the things a community posts constantly: an event, a move-in, a staff spotlight, a testimonial. And one rule that is not negotiable, which is that generated imagery never carries a generated logo, ever.
Then AI does what it is genuinely good at here: backgrounds, textures, variations on a layout that already exists, cleanup, resizing one post into five placements. The judgment about what the brand looks like stays with someone who knows what the building looks like.
A graphic that could belong to the bakery, the handyman, and your competitor is not branding. It is decoration.