Lily Ray's June 2026 analysis of 80 B2B software queries found that AI systems cited self-ranked listicles 323 times. In 224 of those cases (69%), the AI cited the brand's page but recommended a competitor instead. Read that again: the content did the work, earned the citation, and then funneled the buyer somewhere else.
If you're a CMO running a content program that publishes "best X software" roundups with your product conveniently sitting at number one, this is the number that should rearrange your Q3 priorities.
Citations aren't recommendations, and the gap is expensive
The distinction matters more than most teams realize. A citation means the AI treated your page as a source of information. A recommendation means the AI told the buyer to go buy something. Only one of those moves pipeline.
In Ray's data, the brands that earned recommendations had something the cited-but-excluded brands didn't: independent third-party coverage. More referring domains, more mentions on sites the AI treats as trustworthy (Reddit, Forbes, YouTube reviews, comparison posts from unaffiliated creators). The deciding factor wasn't on-page SEO. It was off-page validation.
Think of it this way. LLMs behave like a skeptical buyer who reads your self-promotional listicle, notes what you sell, then checks what everyone else says about you before making a recommendation. If "everyone else" isn't talking about you, the AI recommends whoever they are talking about.
The market has fragmented, which makes this worse
A year ago, you could argue this was a ChatGPT-specific problem. That argument died. ChatGPT still leads AI search at roughly 64–68% market share as of January 2026, but it's declining (down about 19 percentage points year-over-year). Google Gemini grew from 5.7% to 21.5% in the same period. Perplexity grew 370%. DeepSeek and Grok each hold 3–4%.
Optimizing for one AI engine is a single point of failure. Your listicle might cite well in ChatGPT and get ignored entirely in Gemini. The fragmentation means you need visibility across multiple AI platforms, and that visibility comes from the same place: independent, third-party mentions that multiple models can draw from.
Meanwhile, 95% of Americans still use traditional search engines, and 57% prefer them for important purchase decisions. So you can't abandon SEO either. The operating constraint is both/and, not either/or.
Run the audit before you change anything
Before rebuilding your content strategy, figure out where you actually stand. Here's the diagnostic:
- Build your query list. Pull your top 15–20 commercial terms ("best [category] software," "[competitor] alternatives," the BOFU phrases your sales team hears on calls). If you can connect recent call recordings to an LLM via MCP, mine the last 90 days of buyer language for terms you're missing.
- Search each term in ChatGPT, Gemini, and Perplexity. Record two things separately: where you're cited and where you're recommended. Run each query 2–3 times; AI outputs vary.
- Score your share of recommendation, not citation. If you're cited in 12 of 20 queries but recommended in 3, your recommendation rate is 15%. That's your baseline.
- Identify which third-party sources the AI cites when recommending competitors. These are your outreach targets.
The hypothesis: if you secure top-three placement in the specific listicles and roundups that LLMs already cite for your money terms, your recommendation rate will increase because LLMs weight third-party validation over vendor-controlled content. Measure recommendation share monthly. Guardrail: don't let organic search traffic to existing pages drop more than 10% during the shift (that's your stop-loss for over-rotating away from traditional SEO).
The fix is third-party proof, not more vendor content
The operational play is outreach, not publishing. Identify the roundups, comparison posts, and review sites that AI engines cite for your category. Then do the unglamorous work: reach out to those publishers, offer access for honest reviews, build an affiliate program that incentivizes editorial content (not just referral volume), and monitor which partners actually produce credible, indexable coverage.
One compliance note worth flagging: if you're paying for placements in content that reads as independent reviews, the FTC's Consumer Review Rule requires disclosure. Build that into your content ops process before it becomes a problem.
The trade-off is real. This takes longer than publishing another listicle. It requires coordination between content, PR, partnerships, and potentially legal. And it won't move the number this week.
But the alternative is already measured. Sixty-nine percent of the time, your self-promotional content is doing the prospecting for someone else's pipeline. The listicle still ranks. The AI still reads it. And then it sends the buyer next door.
