The AI content doesn't sound like us

This is the most common complaint about AI output, and it almost always has the same cause. Work through these in order — the first one accounts for most of it.

​

1. Your brand description

Everything Ocoya generates is written from your brand profile. A thin description produces thin captions, every time, on every channel. Open Brand and read your description as if you were the AI. Does it say what you sell, who buys it, and how you sound?

We sell things online.

gives the model nothing. Compare:

Acme Co makes hard-wearing outdoor gear for people who would rather be outside. We sell direct to walkers and climbers who care more about kit lasting than about looking new. Plain-spoken, a bit dry, never hyped.

The second produces noticeably different output from the same prompt. If you change one thing, change this. See Set up your brand. Fetch brand next to your website reads your site and drafts this for you — usually a better starting point than a blank field.

​

2. Your visual identity

The same applies to images. Logos and colours are what stop generated designs looking like stock. A brand with no visual identity gets generic imagery no matter how good the prompt.

​

3. Your prompt

Describe the subject, not the style — the style should come from your brand, so you don’t have to restate it every time. Put in what the model can’t know: the product name, the date, the offer, what makes this one different. A post about our new jacket is a weaker brief than the Fell jacket, out Friday, our first fully recycled shell, aimed at people who already own one of ours.

​

4. Tone and length

The composer’s tone (Professional, Friendly, Educational, Bold, Founder-led) and length (short through extra long) change the output meaningfully. If everything reads samey, you may have left both on their defaults for every post.

​

5. Hashtags it invented

If the hashtags are wrong rather than the words, that’s a separate fix: select a hashtag library and the AI is restricted to your tags, with no inventing, translating or pluralising.

​

What not to do

Don’t regenerate repeatedly. Each run costs credits and, with the same brand profile and prompt behind it, tends to produce the same kind of output. Two regenerations without changing anything is usually two wasted credits. Don’t fix it entirely by hand every time. Editing one caption fixes one post; improving the brand description fixes every future one.

​

When it’s close but not right

Generate, then edit. Select the part that’s off and use the AI actions — rewrite, shorten, expand — rather than regenerating the whole thing. See Edit a caption with AI. The post score is also worth opening: a caption that reads flat often fails the same two checks, the opening line and having a concrete takeaway.

​

Related