Can AI be trusted to generate content ideas for Instagram and Facebook?

Short answer: what to trust AI with and what not to

How good are the content ideas AI generates

The answer depends on who is asking.

The Doshi and Hauser study I referred to above contained an interesting detail: the people who gained most from access to AI ideas were the less creative writers. Their texts were rated noticeably higher after AI prompting. For those who already generated ideas well, the gain was minimal.

Translated into SMM terms:

• For a small business owner running the page themselves, AI ideas will most likely be better than whatever they come up with in the evening after work. Here AI raises the floor.
• For an experienced social media specialist AI will not add creativity. It will add speed: what used to take an hour of brainstorming takes ten minutes.

So an individual AI idea is usually fine. The problem appears when you look at all the ideas together.

Why AI post ideas all look the same

The same experiment showed the flip side: texts created with AI ideas were more similar to each other than texts by people. The authors called it a social dilemma: each person individually gains, but collectively everyone produces a narrower range of new work.

For social media the mechanics are very simple. Try asking any AI for “post ideas for a coffee shop”. You will get something along the lines of: behind the scenes of making coffee, meet the barista, a poll about your favourite drink, an interesting fact about coffee, a repost of a guest photo. Every coffee shop in your city that asks the same question gets the same list.

The model offers the most common solution to a typical situation. To get something atypical, you have to give it what is not publicly available:

• the questions customers ask in DMs;
• real reviews and complaints;
• what sells best and what sits on the shelf;
• local context: the neighbourhood, events, the season in your particular city;
• what has already been published and what landed well.

The quality of AI ideas is directly proportional to the quality of what you fed it.

AI for generating Instagram and Facebook content ideas: where it is genuinely strong

There are tasks where AI works better than a human, and there is no point refusing on principle:

• Volume. Fifty topics for a “useful tips” strand in a minute. A person starts repeating themselves after the twentieth.
• One idea, five formats. A 30-second Reels script, a seven-slide carousel, a Stories poll, a post for a Facebook group. This is mechanical work, and AI handles it well.
• Hooks. Ten versions of an opening line for a single post. You will pick two, but without AI you would have had one.
• Repackaging. An old post that did well a year ago can be presented from a different angle, and AI will find that angle faster.
• The structure of your content strands. The balance of educational, entertaining and sales posts across the month, broken down by day. AI builds the skeleton of a content plan well; filling it in is your job. How to put a content plan together systematically is covered in the article “What a content plan is”.

The downsides of AI content for social media

Is it obvious that content was generated by AI

It depends what we are talking about: visuals, or ideas and text.

Visuals get recognised. People easily spot AI images and videos, and Meta labels photos and video created by artificial intelligence on Instagram and Facebook. Mosseri confirms that the platform does not block such content but labels it so users understand where the synthetic material is.

A single text — often not; a whole feed — yes. One post built on an AI idea is hard to tell from an ordinary one. But when the whole month is done that way, the feed becomes flat and predictable, without a single post that grabs anyone. The audience does not articulate that it is AI; they simply stop pausing on your posts. This mass uniformity is exactly what the Science Advances study is about.

How AI texts differ from human ones at the level of words and rhythm is covered in detail in the article “AI copywriting vs a human”.

Why AI content does not convert into sales

The most direct data here came from NielsenIQ. In a neuromarketing study, more than 2,000 participants watched AI-generated advertising ranging from weak to high quality. The result: people easily recognised the AI creative, reacted to it more negatively, and both memory activation and intent to act after viewing were weaker than with traditional advertising.

On top of that, there are two more reasons I see in practice:

• AI does not know your sales arguments. The real prices, the terms, the objections customers raise with your sales team. A generic “Why choose us” post answers not one specific question a buyer has.
• Selling on social media runs on trust in people. The buyer wants to see who is behind the brand: the owner, the team, the customers. The strongest content here is real content from customers, and there is a separate article on its role: “What UGC is”.

Which posts you should not trust AI with

• Posts with prices, promotions and delivery terms. AI will confidently write the wrong figure, and it will go live.
• Reactions to news and events. The model does not know what happened yesterday. A jokey post scheduled via AI can go out on the day of a tragedy.
• Case studies and reviews. A generated “customer review” is a fabrication, and for a brand it ends in lost trust once it comes out.
• The founder’s personal stories. Their value lies precisely in the fact that they happened to a real person.
• The brand’s position on sensitive subjects. War, tragedies, social conflicts — these call for a human decision, not a statistically probable answer.
• Replies to negative comments. A tone AI considers neutral can read to a customer as a brush-off.

I wrote more broadly about which marketing tasks are dangerous to hand to AI without oversight in the article “Why marketing cannot be handed over to AI entirely”.

The model that works: the human sets the frame, AI supplies the volume

The approach that works best with AI at the ideas stage has six steps, and a human is involved in four of them:

The most important step is the sixth, and it is the one skipped most often. If you never feed back into the brief what actually landed with your audience, AI will generate ideas from scratch every month — which means the same generic ones.

In his interview, Mosseri calls the most valuable skill of the AI era knowing what is worth creating in the first place. In social media that is steps 1 and 3: deciding what to talk about, and filtering out what is not worth saying. The rest can be sped up.

Frequently asked questions (FAQ)