What 9 million AI answers reveal about search visibility

AI search visibility doesn’t follow the rules many brands assume. Positive sentiment doesn’t necessarily drive citations, Reddit isn’t the strongest third-party lever, and most commercial-intent citations go to sources outside your own site.

I won’t lie, I used to believe some of these assumptions myself. But instead of relying on gut feel or a single case study built on someone’s randomly selected set of 100 prompts, I looked at the goldmine of data I actually have: 9 million AI answers across nine platforms (ChatGPT, AI Overviews, AI Mode, Gemini, Perplexity, Copilot, Grok, Claude, and Meta AI) for 400+ enterprise brands.

This isn’t a random data set. It’s our enterprise clients and their competitors, all tracked on Writesonic to fuel live AI search campaigns.

Here’s what the data reveals about what actually drives AI search visibility.

1. Positive sentiment doesn’t drive more AI citations

We know AI models are trained on and retrieve language about your brand, so more positive language should mean more citations, right?

Wrong. Across five of the biggest platforms — ChatGPT, AI Mode, Gemini, Copilot, and Grok — sentiment actually correlates negatively with citation frequency. This means brands with higher rates of negative sentiment were, if anything, slightly more likely to get cited.

AI models are optimizing for how often and how deeply a brand gets talked about. So controversy actually creates a bigger “paper” trail, whereas a well-liked brand nobody’s talking about leaves a smaller footprint, for better or worse.

This aligns with research around frequency and LLM knowledge more broadly. A Princeton/Berkeley study showed a model’s ability to surface a fact correlates directly with how many times that topic appeared in its training data.

Key takeaway: Stop treating AI visibility as an outcome of reputation management. Instead, focus on online reputation management (ORM) and generative engine optimization (GEO) as two different workstreams. You should still manage sentiment, of course, as it affects your conversions and your trust, but don’t expect an investment in ORM to immediately move your citation count.

Dig deeper: The AI visibility index: Which brands are vanishing from AI search?

2. Ordinary pages earn most AI citations

Many SEOs I know built their 2025 content calendars around AI-friendly formats. Feed the machine what it likes to eat, right?

Except that’s not what the data shows either. 64% of citations come from ordinary pages. Not listicles, not how-tos, not comparison content. Just normal, thorough, well-written content. 

AI models pull the passage that best answers a prompt, and a genuinely useful product page or detailed explainer often answers more prompts better than a thin, biased “Top 10” post that exists solely to be cited.

Key takeaway: Stop oversimplifying AI optimization as a mere formatting exercise. Make your regular pages more thorough and more useful. This means a focus on better coverage of the questions people actually ask, more specifics, more original data, and more unique subject matter expertise or perspective. It’s the same work that builds value for your users, a stronger SEO footprint, and more conversions.

3. Third-party sources dominate commercial-intent AI queries

It’s your product, your pricing. Who’s a better authority on you than you? Unfortunately, it doesn’t work out that way, and this is where the data gets nuanced.

82% of citations on bottom-of-funnel commercial-intent prompts go to third parties. And owned pages make up just 3%. Such a small share!

Interestingly, bottom-of-funnel prompts are the most impacted this way: informational and navigational queries actually see up to 13% owned coverage. It’s as if AI models decided that when someone’s about to spend money, the credible source is anyone but the seller.

But here’s a twist that should change your roadmap. When your own content does get cited, your AI visibility rate jumps 5x, outperforming even YouTube’s 2.8x lift, the strongest of any third-party source we tracked (and more on that next).

And owned citations last 3-9x longer than third-party ones, which our data showed decay incredibly fast: up to 58% of all citations never reappear after their first showing. So yes, owned content is less cited than third-party content. But it’s the single strongest, longest-lasting lift in the entire dataset.

Key takeaway: Don’t pull back on owned content just because third parties get cited more often. Build and optimize pages specifically around the query types your data shows are already earning owned citations. Make sure to allocate part of your digital PR budget to bottom-of-funnel, commercial-intent topics such as pricing comparisons, category roundups, and “best X for Y” placements.

You can’t win today with just the big-splash data study or C-suite interview. Most PR programs chase awareness moments, but targeted bottom-of-funnel digital PR compounds month over month, and that consistency is what wins you visibility in AI search.

Dig deeper: How to make your SEO strategy more commercially aware

4. YouTube drives the strongest AI citation lift

In a given week, I can pitch two, maybe three enterprise AI search prospects, and Reddit comes up every single time. Licensing deals, LinkedIn posts, and years of conference talks (including some from me) keep telling you that you can’t ignore Reddit. It’s reinforced everywhere you look.

Despite some recent drops in how often Reddit is cited in ChatGPT, Reddit is often a good source to drive search everywhere optimization results. It’s just not the biggest lever.

YouTube actually delivers the strongest lift in AI citations at 2.8x, ahead of LinkedIn (2.7x), Reddit (2.4x), G2 (1.9x), and Gartner (1.6x).

This will differ by vertical, of course, but overall, YouTube is such an underinvested surface in AI search. Most brands treat it as a brand channel or, if they’re sophisticated, a search engine, but not as the AI retrieval mechanism that it is.

But think about what a video transcript actually is: a large, well-structured, spoken-language document. That’s exactly what these systems retrieve.

Key takeaway: Build an AI video roadmap backed by three data sources: YouTube search demand, prompt volume data, and your AI visibility data to see which prompts tied to your audience are already citing YouTube.

Don’t cut corners. You will need full transcripts, optimized descriptive titles, and consistent brand mentions throughout your videos. Based on the campaigns I’ve seen, success comes down to engagement-worthy videos and consistency, so don’t start this if you can’t sustain it.

5. AI citations don’t always build brand visibility

Even once you’re earning the right citations, there’s a catch that should change how you report results. Up to 75% of AI answers that cite one of your pages never actually say your brand name in the answer. This is what’s being called a “ghost citation.”

The model takes your content, your research, your numbers, but unless someone scans or clicks the citation, they may never connect it back to your brand. The user gets all the value, but you may get none of the recognition. So a rising citation count, on its own, can be an empty metric to celebrate.

Key takeaway: Shift your reporting from raw citation counts to visibility and share of voice against competitors. If citation growth (which you can still look at as a leading indicator) isn’t followed by stronger AI visibility, more brand searches, more direct traffic, or more AI referral traffic, those citations might not actually be working for you.

Dig deeper: 7 AI search myths, debunked: What the data says about SEO

6. Brand factors drive more AI visibility volatility than platforms

You’re not alone if your AI visibility jumps around week to week or even month to month. And it can be frustrating to be putting in the work but not be able to sustain results over time. A big assumption I see a lot of people make, though, is that AI models are just inconsistent and unstable, that “it’s just ChatGPT being ChatGPT.”

And while we do know that new models or technical changes can bring shakeups, just like Google algorithm updates always have, our data shows that brands are the variable here, not the platforms.

Across 9 million answers, platform volatility sits in a tight range, 0.08–0.32, while brand volatility spans 0.04-1.00, a range that dwarfs it. On the same platform, in the same week, some brands are rock solid while others swing wildly.

The variable that matters with AI instability isn’t the LLM, it’s the brand. Now, to me, that’s actually good news because that means your volatility is diagnosable and fixable. And based on analyzing the top-performing brands our agency works with, there are a couple of traits they share that I recommend you act on: deep topical coverage and strong third-party mentions.

Key takeaway: When your visibility swings and you’re struggling to sustain meaningful growth, don’t just blame platform inconsistency or instability. While that can be a factor, it’s worth taking the time to audit your owned content, run a competitive authority analysis, and look for where your coverage is thin or your entity is ambiguous. That’s where your volatility is more likely coming from.

AI search still rewards the fundamentals

None of our 9 million data points or recommendations add up to an AI search hack. They point back to the same thing good SEO has always rewarded: being a producer of great content, a trusted authority, and a source worth citing. 

Like SEO, GEO isn’t about gaming a platform. It’s about earning the right to be cited. So anchor your strategy back to those fundamentals, not on whatever small-sample study is trending on LinkedIn this week.