An AI engine can lift a fact straight from your page and, in the same answer, recommend a competitor instead. Being cited and being recommended are two different outcomes — and in 2026 they are visibly decoupling. Two practitioner analyses this year put numbers on the gap and, between them, explain why the popular tactic of publishing your own "we're the best" list now tends to backfire.
What's the difference between cited and recommended?
A citation is when the engine pulls a passage or fact from your page and attributes it as a source. A recommendation is when the engine names your brand as the answer. The first is about being a useful document; the second is about being the trusted choice — and the same answer can do one without the other.
The marketing strategist behind the "GEO is a CMO problem, not an SEO problem" analysis (mid-2026) framed it bluntly: high web rankings and citations do not guarantee the brand gets recommended. In a "best AI SDR agents" query, the brand Coldreach reportedly ranked first and was cited, yet was not the brand the answer recommended. In a "best insider-threat management" query, several brands that had published self-listing roundups were cited but not recommended, while competitors that hadn't gamed the format were the ones named. It's one analyst's reading of specific queries — directional, not a controlled study — but it isolates the gap cleanly.
An AI answer can quote your page and recommend your competitor in the same breath. Winning the citation is not winning the customer — the recommendation goes to the brand the wider web already trusts.
How big is the gap — and why does self-promotion widen it?
Large enough to invert the tactic that created it. SEO analyst Lily Ray's mid-2026 analysis tracked 100 B2B "best [category]" queries between April and June 2026. Her headline finding: when a brand's own self-promoting listicle (one that ranks itself #1) was cited in the AI answer, the self-promoting brand was left out of the actual recommendation roughly 69% of the time — often while competitors named lower on its own page got recommended instead.
A few specifics from that analysis, all attributed to Ray and dated mid-2026 (single-author, treat as directional):
- She identified 184 self-promoting listicle pages across 146 brands, with a sharp spike from 2025 as the "GEO boom" accelerated.
- Examples: Oasis LMS was cited extensively for "best LMS" yet excluded from the recommendation, with Kajabi and Thinkific named instead; Pylon had two self-promo listicles cited while Zendesk and Freshdesk were recommended.
- A January 2026 Google update algorithmically demoted sites leaning heavily on the tactic, with some losses extending domain-wide — not just the listicle pages.
- Google added disclaimers warning users about "self-proclaimed experts" for some queries, and Reddit and Forbes increasingly dominate "best" citations — a tilt toward user-generated and editorial sources.
The mechanism is the one we covered in why AI loves listicles: engines treat a page that ranks only its own author as promotional and discount the recommendation it makes, even while they'll still cite a useful fact from it. So a self-listing page can simultaneously earn a citation and hand the recommendation to the rivals it lists.
What actually earns the recommendation?
Corroboration — what others say about you — not self-assertion. The recommendation tracks the brand the wider web already agrees on, which is why both analyses point to the same levers:
- Be consistent across your own surfaces. Homepage, product pages, and docs should state the same positioning and the same specifics, so an engine builds one coherent picture of you.
- Earn third-party coverage. Reviews, independent roundups, and editorial mentions are the corroboration engines weigh above your own claims — pursue earned placement in lists, per getting into the best-of lists AI cites.
- Build entity strength. The durable signal is how well-established you are as an entity across the web — the slow lever in entity strength, and the one that lets bigger brands still get recommended where smaller ones can't.
- Don't neglect the web ranking. The CMO-problem piece cites research by Kevin Indig finding that web search position has the biggest single impact on LLM citation rate — classic SEO still feeds the system, even as it stops being the whole story.
This is the same conclusion the corpus keeps reaching from different angles — that you can't fake your way past corroboration, and that the content and platform signals behind a pick are separable from the ones behind a citation, which is the heart of why AI cites one brand over a near-identical competitor. For the broader inputs an engine balances, see how AI engines decide which brands to recommend.
What should a brand do differently?
Stop optimising for the citation and start measuring the recommendation. Practically:
- Audit bottom-funnel prompts. For your real "best [category]" and "alternatives to [competitor]" questions, check whether you're recommended — not just whether you're linked somewhere in the sources.
- Retire the self-listing play. If you publish a "best tools" page with yourself at #1, recognise it may be voting for the competitors you list; redirect that effort to earned placement and genuine educational content.
- Own your narrative where you legitimately win. Brand- and feature-specific questions ("how does X work," "X pricing") are where your own pages should be the recommendation, because the engine wants the canonical first-party source.
- Track the gap over time. Citation without recommendation is a measurable, fixable state — but only if you watch both, across every engine.
The uncomfortable takeaway: you can do everything that earns a citation and still not be the brand AI names. Measuring whether you're recommended — not merely cited — across engines and over time is exactly what Buffy Intel is built to do. Questions: [email protected].
