AI Names Your Brand in Only 43% of Citations. Here's Why the Other 57% Stay Silent. [Research]

Imagine writing the definitive guide on customer success for B2B SaaS. Twelve thousand words. Original survey data. A methodology your team coined. A Wikipedia entry. Every checkbox on every GEO checklist, ticked.

Now imagine Perplexity citing that guide in its answer to “what is customer success” and not mentioning your brand once.

Your URL in the footnote. Your ideas in the answer. Your name nowhere.

Experiment A in this series established that 76% of AI citations are syntheses: the AI uses your page as raw material without quoting it directly. Experiment B answers the follow-up question every content marketer needs to face: when AI cites your page, does it even say your name?

The answer, on average, is: not usually. But “on average” hides the most important part of the story, so read past it.

What 1,349 AI Citations Reveal About Getting Named

1,349 real AI citations with an identifiable source brand. 5 AI surfaces. 6 visibility strategies scored. Two experiments combined.

We measured how often brands got named across B2B SaaS questions on Google AI Overviews, ChatGPT, Perplexity, Claude, and Gemini. For each citation where we could identify a known brand as the source, we recorded whether that brand’s name actually appeared in the answer.

We then scored each cited page on six visibility strategies:

• Publishing original data
• Owning a named framework
• Adding a named author
• Having a Wikipedia page
• Being mentioned across the web
• Leading the page with the brand name

Then we ran a statistical model to see which of these actually predicted naming. The model accounted for each page's domain authority, the AI surface, and the type of question asked.

The dataset combines Experiment B (766 rows, early-funnel only) with Experiment C’s early-funnel citations (583 rows), bringing the total to 1,349 rows. The model runs on 1,046 rows with full feature coverage.

What the funnel stages mean. Throughout this piece we use three stages, with the exact prompts we ran:

• Early-stage (TOFU), learning the concept. Example prompt: “what is customer success” / “what is product-led growth”
• Mid-stage (MOFU), comparing options. Example prompt: “best customer success platforms for B2B SaaS” / “best product analytics platforms for SaaS”.
• Late-stage (BOFU), deciding what to buy. Example prompt: “what should I look for when buying customer success software” / “how much does ABM software cost for a 50-person B2B SaaS team”.

The same brand and topic were tested at all three stages, so any difference in naming comes from the question type, not the brand.

Finding 1: The 43% Average Hides Two Completely Different Worlds

Across all 1,349 citations, the brand was named 42.6% of the time. But quoting that average on its own is misleading, and this is the single most important thing to understand about AI visibility today. The 43% is really two separate worlds averaged together.

Note: Ongoing experiment, sample still growing, treat as directional.

There’s a cliff between 96% and 18%. On one hand, ChatGPT and Google’s AI Overviews behave like search engines, when they cite your page, they almost always name your brand. On the other hand, Perplexity, Claude, and Gemini behave like synthesizers, on early “what is X” questions they cite your page and name you less than one time in five.

Why “surface matters most” is the real headline, and why the average misleads. If your buyers mostly use ChatGPT or Google’s AI Overviews, getting cited is almost the same as getting named, at every stage of the funnel. If they mostly use Gemini or Perplexity, a citation usually means your idea is used and your name is dropped. Reporting a single “43% naming rate” to your team implies content tweaks will move it. They mostly won’t, the number you actually get is set first by which AI your buyers use.

A caution on ChatGPT: its ~98% rests on only 55 citations, because ChatGPT cites external sources far less often than the other surfaces. Treat ChatGPT and Google AI Overviews together as the “search-like” group (~96%), with Google AI Overviews, 409 citations, as the solid anchor.

The structural reality: Gemini names the brand on 10% of early-stage citations. No single content strategy in this study pushed that meaningfully higher on its own. You can improve your odds within that ceiling, but the ceiling is set by the surface. Know which AI tools your buyers use before you set any naming-rate target.

Finding 2: Six Strategies Tested. Three Worked. Three Didn’t.

We scored 1,046 cited pages on six strategies you’ll find recommended in virtually every AI-visibility guide published since 2024. The “naming odds” column below translates the statistics into plain language: how much each feature changed a page’s odds of getting the brand named, holding everything else constant.

“Naming odds” is the plain-language version of the model’s odds ratios. The three “Works” results are statistically significant (very unlikely to be chance); the three “No effect” results were not. Exact figures are in the methodology note. The model controls for the page’s domain authority, the AI surface, and the question type.

The three strategies that failed deserve closer attention, because they are the ones most actively promoted.

• Publishing original data did nothing. Pages citing their own survey or proprietary statistics earned brand mentions at the same rate as pages that didn’t. This may reflect timing (new research gets indexed slowly) or a surface effect. Either way, the advice isn’t supported by the data.
• Adding a named author slightly backfired. Pages with prominent human bylines were named less often than corporate pages, not more. The likely reason: when a human author is front and center, the AI credits the idea to the person, not the company. The author gets the credit; the brand doesn’t.
• Having a Wikipedia page made no difference. This is the most striking null result. “Anchor your brand in the knowledge graph” is stated as near-gospel in AI-visibility circles, but once your page is already being cited, having a Wikipedia entry shows no power to make the AI say your name.
• The core distinction: Getting cited is mostly about your domain’s authority. Getting named once cited is about how present your brand is across the web and how your page is structured. Most teams solve the first and assume it automatically solves the second. It doesn’t.

Finding 3: Named Frameworks Work, But Only at Scale

Owning a named framework roughly 1.8×’d the odds of being named, but this signal only showed up in the larger combined dataset. In Experiment B on its own, it looked like nothing.

Here’s what changed. Experiment B was mostly “what is X” and “how to” questions, where pages built around a framework almost never got the brand named on the synthesizer surfaces. Adding Experiment C’s broader question mix, including comparison and “best X” questions, where AI naturally refers to methodologies by name, gave the analysis enough range to detect the effect.

The practical read: owning a named methodology does raise your naming odds, but mainly on the question types where AI naturally invokes frameworks. On pure “what is X” explainers, the framework barely helps. On comparison and best-practice questions, a brand-named framework is a reliable trigger for getting named.

The implication: Don’t just coin a framework. Optimize the page describing it for the questions where AI reaches for frameworks, comparison questions, “best practices for X,” “how leading companies do Y.” That’s where the effect lives.

Note: Because this result appeared only in the combined dataset and wasn’t planned for either experiment individually, treat it as directionally strong but not yet independently confirmed.

Finding 4: Community Pages Name Your Brand Twice as Often as Your Own Content

When we enriched the citation data with source-type classifications, one finding stood out clearly.

Note: Ongoing experiment, sample still growing, treat as directional.

The two big, reliable categories are the ones to anchor on: community pages (123 citations) name the brand 69% of the time, while owned brand pages (1,056 citations) name it 40% of the time. The listicle, earned-media, and encyclopedia rows each rest on fewer than 25 citations, so read them as directional hints, not hard numbers.

Owned brand pages make up 78% of all citations but produce one of the lowest naming rates early in the funnel. Community pages produce the highest.

Why Community Out-Names Your Own Pages

The reason is structural. Owned brand pages are most often cited on “what is X” and “how to” questions (“what is churn rate,” “how to build a customer journey map”), exactly the questions where the synthesizer surfaces almost never name the brand. Community pages tend to show up on comparison questions (“what’s the best CRM for SMB”), where naming rates are naturally higher.

This doesn’t mean owned content is ineffective. It generates the most total citation volume by far. But on a per-citation basis, a Reddit thread discussing your brand converts to a named mention at nearly twice the rate of your own content.

The content strategy implication: Community presence is not just a brand-building play. It is a per-citation naming efficiency play. Community presence earns named mentions that owned content, on early-stage questions, structurally cannot.

Finding 5: The Naming Gap Nearly Closes When People Compare Options

Here is the finding that should change how you benchmark your AI visibility. The naming gap is mostly an early-stage problem. Move from “what is X” to “best X for Y,” and the synthesizer surfaces start naming brands far more often.

The table below recomputes both stages on the same metric and the same dataset, so the comparison is apples-to-apples (the earlier Finding 1 figures came from a slightly different, larger citation set, which is why a couple of numbers differ by a few points).

Here’s the same idea in the actual prompts we ran, for one topic (customer success):

• Early-stage prompt: “what is customer success” - synthesizer surfaces cite a vendor’s guide and rarely name the vendor.
• Mid-stage prompt: “best customer success platforms for B2B SaaS” - the same surfaces now have to list specific options, so brands like Gainsight, Totango, and ChurnZero get named.
• Late-stage prompt: “what should I look for when buying customer success software” - naming stays high; the AI is helping the user evaluate named choices.

The reason is mechanical. “What is X” questions pull the AI into explain-mode, where it summarizes the concept and doesn’t need to credit anyone. “Best X for Y” questions force it into list-mode, where it has to name specific products for the answer to be useful. The question type changes the behavior, and the behavior changes your visibility.

The benchmarking implication: If your AI-visibility dashboard only tracks early-stage “what is X” questions on Claude or Gemini, you’re measuring the floor, not the ceiling. Mid-stage naming rates are roughly 2–10× higher on the synthesizer surfaces. Track comparison questions too before drawing any conclusion about how visible you are.

The Playbook: Earning Brand Names in AI Answers

Step 1: Know which AI tools your buyers use, then set realistic targets.

Find out which AI surfaces your buyers actually use for research. If they lean toward Perplexity, Claude, or Gemini, a 20–25% naming rate on early-stage questions isn’t failure, it’s the baseline the surface allows. Set improvement targets inside that ceiling.

Step 2: Put your brand name in the first sentence of every cited page.

Leading with the brand name raised naming odds by about 66%, a big return on one structural change. Check your top-cited pages now. If the first sentence reads “Customer success is the practice of…” instead of “[Brand]’s approach to customer success…”, you’re leaving that gain on the table.

Step 3: Build mentions across the web, not just great content on your own site.

Being mentioned widely was one of the strongest predictors, roughly 1.7× the odds each time your presence across the web doubles. Every partner page, every G2 review, every Reddit thread, every comparison article that includes you compounds into better naming odds on every citation you earn.

Step 4: Name and describe your frameworks explicitly.

Owning a named framework roughly 1.8×’d the odds of being named in the larger dataset. If you have a proprietary methodology, make sure the page describing it (a) leads with your brand name and the framework name together, and (b) is written for comparison and best-practice questions, not just “what is X” explainers.

Step 5: Track comparison questions, not just “what is X.”

If Claude, Gemini, or Perplexity are in your mix, add “best X for Y” prompts to your tracking. Naming rates are roughly 2–10× higher there, and that’s where your improvement effort will show up first.

Step 6: Invest in community presence as a naming play.

Reddit, Hacker News, and niche forums name the brand ~69% of the time per citation, nearly twice the rate of your own pages. Getting discussed in community threads about your category is one of the highest-efficiency naming moves available.

What This Means for How You Invest in Content

The most common content strategy mistake in AI search is not writing bad content. It is writing good content and measuring the wrong outcome.

Citation volume is what goes in. Brand naming is what comes out. This study makes clear they’re not the same thing, don’t respond to the same fixes, and shouldn’t sit on the same dashboard.

The surface split makes this especially sharp. If your buyers use ChatGPT or Google’s AI Overviews, a citation is almost always a named mention. If they use Gemini or Perplexity, a citation usually means your ideas are used and your name is dropped. On early-stage questions that’s roughly a 9× difference in how often the same page earns a named mention, depending entirely on where it gets cited.

The three strategies that work (lead with your brand name, get mentioned across the web, own a named framework) share one logic: they make your brand name present, repeated, and hard to leave out in the moments where the AI gets to choose whether to say it. That’s the target.

You do not have a citation problem. You have a naming problem. And the levers for fixing it are different.

Limitations and Methodology

This is an observational study combining two experiments in a single collection window (May 2026). The model accounts for each page’s domain authority, the AI surface, the question type, and the brand’s size tier, but we can’t rule out factors we didn’t measure. Observational means we’re describing what correlates with getting named, not proving cause and effect.

The overall 42.6% naming rate blends Experiment B’s original questions with Experiment C’s broader set. Experiment C’s early-stage questions name brands less often (36%) than Experiment B’s (48%), most likely because of different question and surface mixes rather than a real drop in performance.

A note on the two tables: Finding 1’s naming rates come from the full 1,349-citation set, while Finding 5’s early-vs-mid comparison is recomputed on the funnel dataset so both stages use the same metric. That’s why a few early-stage numbers differ by several points between the two tables; they're different citation sets, not a contradiction. ChatGPT’s figures throughout rest on a small number of citations and should be read as directional.

Roughly 60% of what determines whether a brand gets named comes down to which AI surface the question lands on and what type of question it is. The content features explain the remaining ~40%. Strategy matters, but much less than which AI your buyers use.

Data is API-based - Location: United States. Language: English. Single collection window: May 2026.

This is Experiment B in a three-part B2B SaaS AI search study. Experiment A: 76% of AI citations are syntheses. Your ideas without your words. Experiment C: Does early-stage visibility compound downstream? The answer is the most counterintuitive finding in the series.