Why Does AI Keep Recommending the Same Brands? Here’s What New Research Shows

Two new studies published in late July 2026 expose a visibility gap most brands aren’t measuring, and the fix starts long before any query gets typed.

Here’s a question worth sitting with: when someone asks ChatGPT or Google AI Overviews to recommend a product in your category, is your brand even in the running?

Not in the “our page ranks well” sense. In the deeper sense, whether the AI model actually knows who you are before it starts searching.

Two pieces of research published in the last few days make this uncomfortably concrete. The first, from geoSurge, found that AI models search for brands they already know at 3.2 times the rate of brands they don’t. The second, from Writesonic analyzed on Search Engine Land, found that roughly 40% of AI citations don’t even name the brand being cited. You can be the source and still be invisible to the reader.

Both findings point to the same shift: AI search visibility is not primarily a content or keyword problem. It’s a brand familiarity problem. And the work that fixes it looks different from anything most SEO playbooks have covered before.

How AI Recommendations Actually Work

When someone asks an AI model to recommend a software tool, a financial product, or a marketing agency, the model doesn’t start from a blank slate. It arrives with a mental shortlist already formed from its training data. That shortlist reflects which brands appeared most consistently, most authoritatively, and most frequently in the content the model learned from.

Only after that does the model run web searches to verify and fill gaps, a process called query fan-out. Those fan-out searches tend to be a mix of generic category queries and specific brand name lookups. Here’s the key insight: the brands it searches for by name are overwhelmingly the ones already on its internal shortlist.

The geoSurge research team measured this directly. They tested 66 buyer-style prompts across nine industries, running each prompt 60 times across a 12-day window from late May to early June 2026. They analyzed nearly 4,000 model responses and more than 13,000 fan-out search queries. The finding was consistent across every industry tested: brands the model already remembered were searched 3.2 times more often than brands it didn’t recognize. Among the top-5 recalled brands, the search rate reached 67%.

That’s not a small edge. It means entering an AI recommendation round without established model memory is like competing in a tender process where most brands were already shortlisted and you’re starting from scratch every single time.

The Ghost Citation Problem

If the geoSurge finding is about getting into the AI’s consideration set, the Writesonic finding is about getting credit once you’re there. According to analysis of roughly 16 million brand appearances published on Search Engine Land in late July 2026, around 40% of AI citations don’t name the source brand in the generated answer.

That means the AI uses your page as a source, links to it in the footnotes, and then presents the information without telling the reader it came from you. The URL is there. Your brand name isn’t.

Ghost citation rates varied by platform. Perplexity dropped the brand name in 52% of citations. Google AI Mode reached 49%, and Google AI Overviews hit 41%. ChatGPT came in at 37%. Gemini and Microsoft Copilot were lower, at 25% and 19% respectively.

This creates a second visibility layer most marketers aren’t tracking. Your site can appear in citation reports, and your AI search numbers can look healthy on paper, while the actual readers of those AI answers have no idea who produced the information they just read.

Building AI Brand Memory: The Real Strategy

Model memory is not formed at query time. It’s earned over time through the content and signals that existed when the model was trained and updated. That changes the urgency a bit, but it also changes what you should be doing right now to influence where the next wave of AI training puts your brand.

Earn category association through third-party coverage

The geoSurge team put it plainly: the work that builds model memory is sustained category authority, the kind earned through analyst and press coverage, partnership signals, and consistent association between your brand and its category in the comparison and listicle content that AI models learn from.

If your brand appears in five “best CRM for small teams” roundups, three analyst reports, and a handful of industry case studies, the model develops an association between your brand and that category. If you’re absent from all of those and only visible on your own website, you’re building a house AI will never visit.

Strong off-page SEO is no longer just a ranking tactic. It’s the mechanism by which your brand enters the training data of future AI models. Guest features, earned media placements, PR around named research, and industry mentions all contribute directly to this.

Create named research and frameworks

The ghost citation research from Writesonic points to something useful here. Named research is harder to ghost than generic advice. When a study or framework carries your brand’s name, the data and the brand become harder to separate. A stat attributed to “The Incisive Growth Performance Benchmark” is harder for an AI to cite without naming you than an unnamed tip pulled from a blog post.

This is one of the highest-ROI content investments a mid-size agency or brand can make right now. Original data, even a simple survey or benchmark, paired with a named methodology, gives journalists a hook to cover you, creates third-party citations that compound into AI training data, and tends to be repeated verbatim in ways that keep your brand name attached.

Get explicit with attribution on your pages

One likely factor in ghost citations is how closely your brand name sits to the specific claims on your page. A stat that appears several paragraphs away from the company name that produced it is easier for an AI to lift and anonymize. Pair the attribution directly with the claim: “Incisive Growth’s 2026 audit data showed that…” not just “Our data showed that…” buried three paragraphs after a heading with your logo at the top.

Practical content optimization for AI search means restructuring key pages so your brand name appears alongside every proprietary insight, not just in the header or footer.

What Industry Matters More Than You Think

One underreported finding from the geoSurge study is how much the industry context shapes the model’s behavior. In automotive and finance, 77% to 82% of brand-specific fan-out searches named one of the model’s top-5 recalled brands. In fitness and wellness, only 50% did. The model searches past its own memory more often in less consolidated categories.

The practical implication: if you’re in a relatively consolidated category where a handful of established brands dominate awareness, getting into model memory is close to a prerequisite for showing up at all. In a more fragmented category, strong live content can still pull you into fan-out searches even before your recall is established. But model memory remains the more durable position in either case, since it doesn’t depend on being re-discovered on every query.

Common Mistakes to Avoid

Measuring AI citations without measuring brand mentions

If your AI visibility reporting only tracks whether you appear as a source, you’re missing whether readers ever see your name. Track mentions and citations as separate metrics. A citation rate that looks healthy can hide a ghost citation problem that’s costing you brand recognition with every answer delivered.

Relying only on your own site to build AI presence

Your website’s content contributes to model memory, but the training data AI models learn from is dominated by third-party signals: review sites, comparison pages, media coverage, industry directories. Brands that invest heavily in their own site while neglecting earned presence elsewhere are building a one-legged stool for AI visibility.

Treating this as an SEO-only problem

The geoSurge data makes clear that what determines AI recommendation behavior is more closely related to brand authority than to keyword optimization. Traditional SEO practices still matter for the live content that gets surfaced in fan-out searches. But model memory is built differently, through PR, partnerships, consistent third-party mentions, and the kind of category authority that earns you a place in training data. Marketing teams and SEO teams need to be working this problem together.

What This Means Going Forward

The shift from traditional search to AI-mediated recommendations creates a visibility funnel with two failure modes. You can fail to enter the model’s consideration set at all, the memory problem. Or you can be cited but never named, the ghost citation problem. Either way, the customer never connects your brand to the answer they just received.

The brands that will hold and grow visibility in AI search over the next two to three years are the ones investing now in the category authority that feeds model memory: original research, consistent earned media, strong third-party presence, and content structure that keeps your brand name attached to your ideas.

At Incisive Growth, this is increasingly the lens we apply when auditing a client’s search strategy. Ranking well in traditional search results is still valuable, but it’s no longer a reliable proxy for whether AI models know who you are or will recommend you to a buyer in your category.

If you’re not sure where you stand on either front, that’s the right question to be asking right now, not after the next wave of AI model training has already locked in someone else’s category position.

Frequently Asked Questions