Latest / Social Media Marketing with Fexingo: Instagram, TikTok, LinkedIn, and Platform Strategy / How Brands Use AI to Write Social Media Captions
Transcript
- Lucas: So there is this growing trend where brands are using generative AI to write their social media captions — like, full-on, from scratch, publish-ready captions for Instagram, TikTok, LinkedIn. And the data suggests it might actually work better than a human writer in some cases. Luna: Wait, better? I thought the whole pushback on AI content was that it sounds generic and kills brand voice. What data are we talking about? Lucas: Good question. So the marketing analytics platform Emplifi ran a study earlier this year — they looked at over 10,000 brand posts across Instagram, TikTok, and LinkedIn. They compared posts where the caption was fully human-written versus posts where the brand used an AI tool — mostly GPT-4o or similar models — to either draft or heavily edit the caption. Luna: And what did they find? Did the AI captions tank engagement? Lucas: Actually, no. On average, ai assisted captions saw a 6 percent higher engagement rate on Instagram and a 4 percent higher rate on LinkedIn. TikTok was basically flat — no significant difference. But here is the kicker: the brands that used AI cut their caption production time by about 40 percent. So they were producing more content, faster, with no loss in performance. Luna: Okay, 40 percent time savings is huge. But I'm still skeptical about brand voice. If every brand uses the same model, don't captions start to sound the same? Lucas: That is the fear, and honestly, if you just copy-paste the raw output, yeah, they do. But the brands that are seeing success are not doing that. They are using AI as a first draft engine, then layering in their own voice, inside jokes, specific references. One brand I looked at — a mid-sized consumer goods company that sells organic snacks — they set up a system where the AI gets a prompt that includes their brand guidelines, their tone descriptors, and even examples of past high-performing captions. Luna: So it is more like a smart assistant than a replacement. That makes sense. But how do they measure whether the caption is actually good before posting? Lucas: That is where the interesting new tooling comes in. There are now platforms like Copy.ai and Jasper that integrate directly with Instagram and TikTok APIs. You generate five to ten caption variations, then you can A/B test them — or actually, what some brands do is they run a quick sentiment analysis on each option using a separate AI model. They pick the one with the highest predicted engagement. Luna: So it's AI judging AI? That feels a bit circular. Lucas: It can be, if you are not careful. But the better approach is to combine AI prediction with human judgment. For example, the snack brand I mentioned — they have a human community manager who reviews the top three ai chosen captions and makes the final call. The AI saves them from staring at a blank page, but the human keeps the soul. Luna: I want to dig into the LinkedIn versus TikTok difference. Why would LinkedIn be more receptive to AI captions? My guess is that LinkedIn is more text-heavy and professional, so structure matters more than personality. Lucas: Exactly. LinkedIn users are scanning for insights, not entertainment. An ai generated caption that is well-structured, uses bullet points, and ends with a question — that performs really well. On TikTok, the caption is almost secondary to the video — it's often just a few words or a hook. So the AI advantage is smaller. Luna: And Instagram sits in the middle — still visual-first, but captions can be longer and more narrative. That tracks. Lucas: Right. And the Emplifi data confirmed that: the biggest lift from AI was on LinkedIn, then Instagram, then TikTok. But interestingly, on TikTok, ai generated captions that included emojis and slang actually performed slightly worse than human-written ones. The AI was trying too hard to sound young. Luna: There's a lesson there. You can't fake authenticity. Even if the AI can mimic slang, the audience can smell it. Lucas: Exactly. So the best practice that is emerging is to use AI for the heavy lifting — generating multiple options, structuring long-form captions, optimizing for keywords — but to have a human editor who knows the brand's voice do a final pass. And maybe even run a 'humanness' check. Luna: A humanness check? What does that look like in practice? Lucas: Some brands are using a separate AI tool to score how human a caption sounds — metrics like sentence length variance, use of contractions, personal pronouns, even the frequency of emojis. If the score is too low, they rewrite it. Luna: So AI grading AI for humanness. That is a strange loop, but if it works... I guess the takeaway is that AI is a tool, not a replacement, and the brands that treat it as such are winning. Lucas: Exactly. And the numbers back it up. The snack brand I mentioned — they went from publishing 50 posts per week to 80 posts per week after adopting AI captioning. Their engagement per post stayed the same, but total engagement went up because they were posting more. And their community manager said she spends less time writing and more time actually engaging with comments. Luna: That is a real workflow improvement. I wonder how this scales for smaller businesses that can't afford a dedicated community manager though. Lucas: That is a fair point. For a solo entrepreneur or a two-person marketing team, AI captioning can be a lifeline. There are tools like Writesonic and Simplified that are built specifically for small teams — they generate captions, schedule posts, and even suggest hashtags. The quality is not as high as a custom-tailored system, but it gets you 80 percent of the way there. Luna: And 80 percent is often good enough when you are just trying to maintain a presence. But for brand-building, you probably need that human touch. Lucas: I think that's the right balance. Use AI to maintain volume and consistency, then invest human creativity in the posts that matter most — product launches, campaigns, moments where brand voice really counts. Luna: So if I'm a brand listening to this, what should I do differently starting tomorrow? Lucas: Start small. Pick one platform — probably Instagram or LinkedIn — and try using an AI tool to draft captions for one week. But set a rule: you have to rewrite at least 30 percent of the AI output. That forces you to inject voice. Measure engagement compared to the previous week. If you see a lift, scale it. If not, adjust your prompts or go back to human-only for that platform. Luna: I like that. Quick, low-risk experiment. And I bet most brands will find that their engagement stays the same or improves, while they save hours per week. Lucas: That has been the pattern for the brands I've studied. The ones that fail are the ones that treat AI as a set-it-and-forget-it solution. The ones that succeed treat it as a collaborator. Luna: Alright, next time you come in, I want to talk about ai generated images for social media — that is a whole other can of worms. Lucas: That's a great idea. There's a lot happening with tools like Midjourney and dall e 3 being used for Instagram stories and TikTok backgrounds. We'll get into that soon. Luna: Looking forward to it. For now, I think the key takeaway is: AI captions can save you time without hurting engagement, but only if you keep the human in the loop. Lucas: Couldn't have said it better myself. And on that note, we'll see you next time.