What actually makes AI recommend one brand over another?
by•
Two brands can offer similar products, target the same audience, and have equally strong websites, yet one consistently appears in AI-generated recommendations while the other barely gets mentioned.
We've been exploring how factors beyond a brand's own website, including third-party mentions, reviews, comparisons, and category authority, influence these recommendations.
With ReSO AI, we're working on helping brands understand what's driving these differences and where they can improve.
What do you think carries more weight in AI recommendations: a brand's own content or what others say about it?
114 views
Replies
I think what others say about a brand carries more weight. Anyone can write nice things on their own site, but when the same brand keeps showing up in reviews and comparisons, it feels more believable. Own content still matters, just as the base..
@kate_ragotte19 Agreed, we got more SEO for our product from references on Social Media and customer reviews, than anything else so far.
@kate_ragotte19 Agreed. A brand's native quality of product and service will always triumph long term. But what if there are multiple brands with little to no differentiation between the net reviews or sentiment on third party forums. An LLM has limited numbers to show. That's where on-site efforts kick in.
I think it's the social proof across the entire internet that weighs more heavily and all the referral sites with "follow = yes" More detailed analytics are needed now though for brands - especially upcoming startups @swati_paliwal
@swati_paliwal @sudo_tallan
Yeah so the comment above lays it out nicely. Different models cite differently
Reddit and Wikipedia are always present on top 3 most cited domains across a run (if you were to combine cumulative citations across ChatGPT, Google & Perplexity.
LinkedIn/Youtube breaches top 10 depending on the industry.
Social proof is always important. Where SEO/AEO comes in, is that if 2 brands have relatively same social proof (or within range), the one with better AEO efforts will come up and get more revenue
@swati_paliwal @sudo_tallan @mohit_gupta138 Is AEO a real thing are there quick wins to gain in that space
@swati_paliwal @sudo_tallan @mohit_gupta138 @mertcansaglam Well, our ethos is incremental positive changes done consistently.
There are no quick wins. But there are definitely quick loss-avoids ;P
I wouldnt want to say on a public forum that abc quick fix will get you a boost. But there are definitely highly critical blockers that websites might have that will never let you get cited (e.g. blanket blockers on crawlers)
We measured this last month: 960 answers from ChatGPT, Gemini, Claude and Perplexity to 48 "which X is best" questions, with web search on. The strongest signal was how many third-party pages in Google's top 10 mention the brand. Brands with zero such mentions got recommended in about 19% of relevant answers, with 2–3 pages 34%, with 6+ pages 71%. Brand fame is a confounder, so it's a correlation, not proof.
Sources differ a lot by model too: Claude leaned on "best X" roundups (78% of its citations), ChatGPT more on brands' own sites (62%), and Gemini was the only one citing Reddit and YouTube noticeably. So for the question you asked: what others say, and specifically the roundups that already rank in Google.
@ksanyokm Largely agreed. These also change with model selected for each LLM.
On our side we both create "Best of" content and reach out to third party sites for inclusion
@mohit_gupta138 Makes sense, doing both covers the two source types we saw. One thing from our data: the roundups that correlated with recommendations were the ones already ranking in Google's top 10 for the question, not just any "best of" page.
So for outreach we'd rank targets by their current position first. How do you measure whether a new inclusion actually moved the model's answers? Re-running the same prompts after the page gets indexed?
@mohit_gupta138 @ksanyokm Largely yes. But not on a single run. LLMs have variance baked in. So one-off citations should be taken with a grain of salt. We keep a lookout on weekly and monthly trends. The citation should grow across prompts as well as engines for it to be validated.
@mohit_gupta138 @mohit_resollm Agreed, single runs are too noisy to judge anything. That's why we looked at 960 answers across 48 questions instead of spot checks. Growth across prompts and engines over weeks sounds like the right bar. Thanks for sharing how you validate it.
Both, but they do different jobs. What others say gets you into the answer. Your own wording decides how you're described once you're in. We tested it on our own site: a page had one description of the company in the body and a different one in the footer, and the model quoted the body line both times. So third party mentions get you named, but your own pages still write the sentence.
@lancelot_d_souza1 Fair. But keep in mind that even though the LLM mentioned the body line, it still might have parsed the footer line. And LLMs prefer entity standardisation. It not only helps take control of narrative but also helps with visibility.