Sunday, 10 p.m. You open the chat and type: "give me 10 post ideas about personal finance." Seconds later you get "5 mistakes keeping you from saving money," "How to build an emergency fund" and "Myths about investing." Then you scroll your feed and find three profiles in your niche with posts almost identical to that list.
The problem starts with the request. To use AI for social media content research without landing on generic ideas, give it real material: posts from your references, your own posts, this week's news and the questions your audience keeps asking. Then ask it to read and compare. Ask it to invent, and it invents the same thing for everyone.
Why AI gives you the same idea it gives your competitor
Without specific material, a language model answers with whatever is most likely for the topic. Most likely means most said. Your competitor types the same request and gets the same average of the internet.
Talent and prompt wording have nothing to do with it, and it still costs you. You spend the night rewriting to strip out the AI feel, your brand voice disappears along the way, and the final post has nothing only you could have written. That part that's only yours is what makes someone see you as an authority on the subject.
Generic posts start in research, before the first line is written
Plenty of people try to fix in the caption a problem that got in much earlier, when the subject was chosen. That's why swapping the prompt doesn't rescue the topic. You can ask for "be creative," "avoid the obvious" and "write like an expert." Without material, AI keeps drawing from the most common answers, now with better adjectives.
What changes the result is the kind of request: asking AI for an opinion or asking it to read material.
| Request that produces generic | Request that produces something yours |
|---|---|
| Give me post ideas about sports nutrition | Here are the captions from the last 20 posts of 3 reference profiles. Which topics do they all repeat, and which question does none of them answer? |
| Write a post about real estate market trends | Here is yesterday's news article about [topic]. What angle would a real estate agent with 10 years of new-development launches take that a news portal wouldn't? |
| Create a viral carousel | This carousel got far more comments than the others on the profile. What is the hook, what is the structure and why did it work? |
| Write in my voice | Here are 15 of my captions. List my expressions, my rhythm and how I address my audience. |
AI reads what you hand it very well. When it has to guess, it hands back the average.
The five materials to hand over to AI
- Posts from your niche references. Pick 2 or 3 profiles and copy what they actually published. In videos, the subject is usually in the spoken words, so bring the transcript along with the caption. Ask which topics repeat, which hook patterns show up and which formats dominate. The step-by-step for collecting this is in how to analyze a competitor profile with AI.
- Your own posts. Set aside the ones that sparked the most conversation and a few that went unnoticed. Ask what the best ones have in common, which topics you've already covered and how you really talk. To choose which ones to use, see how to find your best-performing content.
- This week's industry news. Use the specialized outlet, the industry newsletter or the rule change everyone is talking about. Paste the full article text along with the headline. Ask what authority angle your brand would take and what changes for your audience.
- Real audience questions. These are the comments, the story question-box replies and the DMs that keep repeating. Copy them by hand and paste everything in one block. Ask which questions come up most and which ones you haven't answered in any post yet. There's more on this in content ideas.
- Your documented knowledge. Classes, PDFs, presentations and client-session notes hold what only you know. Ask which ideas in that material disagree with what the niche usually says and which cases would make a post.
Check what you have before asking for topics
If you have none of these materials on hand, it's not time to ask AI for topics yet.
Ask AI for the mechanism of the post that worked
- Hook: the first sentence or the first slide, and what it promises.
- Structure: the order in which the arguments appear.
- Format: carousel, talking Reels, split screen, list.
- Tone: provocative, didactic, serious, funny.
- Why it works: what makes the audience stay to the end and comment.
- Audience belief: what most people believe.
- Contradiction: the slide that says this is wrong.
- Practical proof: a case that backs up the contradiction.
- Application: what to do from now on.
- Belief: "Freelancers don't need to worry about taxes."
- Contradiction: she shows why that idea is wrong.
- Proof: a case from her own experience.
- Application: what a freelancer should check.
Never ask for "rewrite this post in other words"
The result is someone else's post with synonyms.
Ask questions that force AI to take a side
Which topic do these profiles repeat that nobody contradicts?
Which audience question shows up in the comments that none of these posts answers well?
Where does my experience, as described in this material, disagree with what these profiles say?
Which of these topics have I already covered, and which have I never touched?
From this news story, what changes in practice for [audience] over the next few weeks?
Does this trending subject have a real connection to my audience, or is it just trending?
List 3 topics that would only make sense on my profile, and explain which material each one is based on.The last question is the most useful. When AI has to say where each idea came from, the generic topic shows up right away, because it points to no material.
The second to last works as a filter for trends. A trending subject only becomes a topic when it has a real connection to your audience and you have something to say about it. If AI has to force the link to your niche, drop it.
Use your real voice as a filter before publishing
Even with good research, the text can come out with AI habits. Paste 10 to 15 of your captions or transcripts and ask for the expressions you repeat, the rhythm of your sentences, how you address your audience and what you would never say.
Then look for these signs in the draft:
- an opening with a rhetorical question;
- mirrored sentences, like "it's not X, it's Y";
- a motivational closer;
- three adjectives in a row;
- a CTA that would fit any profile.
Finally, run the @ test. Imagine the post published on another person's profile in your niche. If it still makes sense there, go back to the research. Editing only the text won't fix it.
In Creathoria, Post personality reads the real text of the brand's posts and extracts voice, rhythm, expressions, catchphrases and the way it addresses the audience, and that goes into every piece. Pieces also go through automatic quality checks and a human-voice reviewer. If the two conflict, the briefing's tone of voice always wins.
How this method works inside Creathoria
- Brand briefing: niche, audience, experience and credentials, offer, tone of voice and forbidden topics.
- Reference profiles: you add the @ and choose which posts to analyze. The AI reads images, carousels slide by slide and video (caption or speech transcript) and builds a dossier with themes, hook patterns, structures, formats, tone and a summary of each post. You can chat with the dossier in the profile chat.
- Own profile analysis: the same study applied to your brand, with what worked, your real tone and the themes already covered.
- Knowledge base: PDF, DOCX, TXT, MD, links and notes become cards.
- Radars: the Reference Radar summarizes what your references posted and suggests hooks to test. The News Radar brings topics with an authority angle from industry news, with a "Go deeper" option. The Trends Radar looks for trending subjects on Google Trends, X and TikTok and filters them for your niche.
- Planning: automatic research uses the dossiers, the knowledge base and your audience's questions. Each topic comes out with a theme, audience question, hook, structure, format and goal.
In every post, Post sources shows which material it was based on: references, news, trends or cards. It's the "demand the source" step from the questions section, applied to every topic.
Instagram profile analysis is on every plan. TikTok, LinkedIn and the radars start on the Creator plan.
Frequently asked questions
Can you use ChatGPT to research social media content?
How do I keep AI-written text from sounding robotic?
Isn't analyzing a competitor's post with AI just copying?
Are the trends AI suggests worth using?
5-minute checklist before sending a topic to production
- Do I know which material this topic is based on?
- Does it answer a real question from my audience?
- Did I take only the mechanism from the reference, with none of the author's sentences?
- Is there something here that comes from my experience or my material?
- Does the text pass the @ test?
- If there's a trend, does it have a real connection to my niche?
Next time you open the chat, paste the captions, the news story and your audience's questions first. The topic only you can post is in that material.
Add your brand briefing and one reference profile during the 14-day trial. Then open the Post sources of the first topic and see which material it was based on.