A social media strategy for AI search structures content so AI engines cite it, not so audiences engage with it. The four moves: baseline which sources already answer for your category, earn placement on third-party domains, publish in Q&A and comparison formats, and maintain your back catalog. Brands are 6.5x more likely to be cited through third-party sources than their own.

‍

The Report That Makes Your Social Dashboard Look Wrong

Open your social analytics. Reach, engagement rate, follower growth. All trending the right way.

Now open ChatGPT and ask it to recommend an agency in your category. Watch it name three competitors and not you.

Both things are true at once, and that gap is the whole problem. Your social strategy is succeeding at the job it was designed for in 2019 and failing at the one that now decides whether buyers ever hear your name.

Here’s the number that explains it. AirOps analyzed 21,311 brand mentions across 500+ commercial-intent queries on GPT-5, Claude Sonnet 4.5, and Perplexity Sonar. 85% of brand mentions came from external domains. Just 13.2% came from the brand’s own site (1).

Which means 85% of your AI visibility is decided on property you don’t control,  and most social strategies spend 100% of their effort on the 15% you do.

That’s the 85/15 problem. Four moves fix it.

‍

Move 1: Find out who is already answering for you

You cannot fix a visibility gap you haven’t measured. Before changing a single post, build a prompt baseline.

Write 15–20 prompts a real buyer would type. Not keywords — questions. “Best social media agency for nonprofits in Oakland.” “How do I choose a wealth management marketing partner?” “[Competitor] vs [you].”

Run each one in ChatGPT, Google AI Mode, Perplexity, and Gemini. For every prompt, log three things: were you mentioned, were you linked, and which domain got the credit instead.

That third column is your strategy document. If a Reddit thread from 2023 and a YouTube comparison video are answering for your category, you now know exactly where to show up, and it isn’t your own feed.

Re-run the identical prompts at 30, 60, and 90 days. This is the only honest way to report progress, because it measures what you’re actually trying to change.

Do not expect AI citations within month one. This is a compounding system, not a campaign. Any agency or tool telling you otherwise is selling you a number it cannot control. The baseline is what protects you.

‍

Move 2: Borrow the domains that already get cited

Of the third-party mentions AirOps tracked, nearly 90% came from listicles, comparisons, or reviews. Not from brand accounts. From other people’s content (1).

So the highest-return social activity is frequently not publishing. It’s earning placement:

  1. Get into “best X for Y” roundups in your category. One inclusion in a well-cited listicle can outperform a quarter of owned posting.
  2. Answer in comparison threads where your category is genuinely being discussed. A useful, specific reply on someone else’s thread lives on a domain AI engines already trust.
  3. Build review-platform presence. G2, Yelp, and similar surface heavily in recommendation queries.
  4. Appear on other people’s channels. Guest spots on YouTube, podcasts, partner webinars.

Uncomfortable, and that’s the point. This move asks your team to spend time on content with someone else’s logo on it. The data says that’s where the citations live right now.

‍

Move 3: Shape content the way engines extract it

Semrush analyzed 217,000 prompts and 248,000 cited Reddit URLs across Google AI Mode, Perplexity, and ChatGPT Search (2). Format was the dominant driver:

What gets citedFinding
Q&A threadsOver half of all citations
Q&A + comparison + discussionNearly three-quarters combined
Median length of a cited post~80 words
Median upvotes on a cited post5–8
Cited posts with under 20 upvotes80%

Read those last two rows again. Eighty percent of cited posts had fewer than 20 upvotes. Engagement is not the mechanism. Relevance and structure are.

For a lot of social teams, this is genuinely good news. You can stop chasing reach and start chasing specificity. A nine-upvote post that answers one question cleanly beats a nine-hundred-upvote post that’s entertaining and vague.

The working test: for every piece of content, name the single question it answers. Then check whether the answer survives being pulled out of its context. If it only makes sense inside the thread, an engine can’t use it.

‍

Move 4: Hold what you’ve built across more than one platform

Two findings that have to be read together.

Your archive is an asset. Cited Reddit content averaged roughly 900 days old, nearly two and a half years. AI engines favor durable answers over fresh ones. So audit your strongest evergreen posts from the last three years, update dead figures and links, and stop deleting old content to tidy a profile.

But concentration is fragile. Semrush tracked 230,000 prompts and over 100 million citations from July to October 2025 (3). Reddit appeared in roughly 60% of ChatGPT responses pre-September, then around 10% post-September: one platform, one month, a six-fold drop.

And most brands have no hedge: AirOps found 68% appeared in only one AI search platform.

Peec AI’s analysis of 30 million sources across five engines ranked Reddit first, YouTube second, LinkedIn top five, with real differences between them (4). ChatGPT leaned Wikipedia and Reddit; Google leaned Facebook and Yelp; Perplexity leaned Reddit, LinkedIn, and G2 for B2B.

Pick a primary and a hedge. B2B: LinkedIn primary, Reddit or YouTube hedge. Then make the profiles themselves readable. Instagram professional-account public posts have been Google-indexable since 10 July 2025 (check the opt-out isn’t on), and Google Search Console platform properties for Instagram, TikTok, X, and YouTube reached general availability on 29 July 2026, giving you real query data to feed back into planning.

‍

“But engagement is what people are looking for”

Fair, and you don’t have to abandon it. Add a second scoreboard rather than replacing the first.

Keep reportingAdd
Follower growthPrompt visibility rate: % of your 20 baseline prompts mentioning you
Engagement rateCitation rate: % that link you
ReachThird-party mention share: mentions on domains you don’t own
Post frequencyPlatform coverage: how many AI engines mention you at all

The second column will be in every client’s QBR deck within a year. Building it now costs you an afternoon.

‍

Start With The Baseline

You need to know where you stand.

Block ninety minutes this week. Write your 20 prompts, run them across four engines, and fill in that third column: which domain got the credit. Most teams find the answer uncomfortable and immediately actionable.

Want some help? Schedule your complimentary session. No commitment, so you can get started on the baseline and consider taking a step forward to create a tailored strategy based on your needs.