How LLMs Construct Your Brand Reputation Without Asking You
When a large language model is asked about your brand, it does not check your website. It draws on a compressed version of everything it was trained on: reviews, forum threads, press coverage, competitor comparisons, analyst commentary, social discourse. The picture it builds is not your brand as you intended it. It is your brand as the internet recorded it.
That distinction matters more than most marketing teams currently appreciate. LLMs are becoming a primary discovery layer for buyers doing early-stage research, and the brand signal those models surface is almost entirely outside your direct control. Almost.
Key Takeaways
- LLMs construct brand perception from third-party signals, not owned content. Your website copy is largely irrelevant to how a model describes you.
- The brands that fare best in AI-generated responses are those with consistent, specific, externally validated positioning accumulated over time.
- Vague brand positioning is penalised in LLM outputs. Models default to category generics when a brand has not earned a distinct signal.
- Earned media, review volume, and consistent brand voice across third-party sources are now brand infrastructure, not just communications tactics.
- You cannot optimise your way into a strong LLM presence in six weeks. The inputs that shape it take months or years to accumulate.
In This Article
I have spent more than two decades watching brands invest heavily in owned channels and then wonder why perception does not shift. The same structural problem applies here, only the stakes are higher because the output is a confident, synthesised answer rather than a search result the user has to evaluate themselves.
What LLMs Actually Train On
To understand how a model sees your brand, you need a basic working model of what goes into training data. LLMs are trained on large corpora of text from across the web: news articles, blog posts, Reddit threads, product reviews, Wikipedia entries, LinkedIn posts, academic papers, press releases, earnings transcripts, and more. The weighting and curation varies by model, but the principle is consistent. Frequency, authority, and consistency of signal all influence how a brand is represented in the model’s internal understanding.
Your homepage copy, your brand guidelines, your carefully crafted mission statement: these are not absent, but they carry far less weight than the aggregate of what third parties have written about you. A single well-placed press piece in a high-authority publication carries more signal than a hundred pages of polished owned content. A pattern of consistent customer reviews over three years shapes the model’s understanding of your service quality more than any claims page on your site.
This is uncomfortable for marketers who have spent years treating owned channels as the primary vehicle for brand building. It is also clarifying. The question is not “what does our website say?” but “what has the world said about us, and how consistently?”
Brand strategy has always required this kind of external orientation. If you want a broader framework for thinking about positioning, differentiation, and how brands earn their place in a category, the Brand Positioning and Archetypes hub covers the strategic foundations that underpin all of this.
Why Vague Positioning Gets Punished
I spent several years judging the Effie Awards, which meant reviewing hundreds of campaign submissions and, more usefully, the strategic briefs behind them. One pattern I saw repeatedly was brands that had invested significantly in creative production but had never resolved a fundamental positioning question: what do we actually stand for that no one else can claim?
LLMs surface this problem with brutal efficiency. When a model is asked to describe a brand and cannot find a clear, consistent, externally validated signal about what makes that brand distinct, it defaults to category-level generics. Ask an LLM about a mid-market B2B software company with vague positioning and it will describe features, mention the category, and offer nothing that differentiates. Ask it about a brand that has earned a specific, consistent reputation over time and it will reflect that specificity back.
This is not an AI problem. It is a positioning problem that AI has made visible. Existing brand building strategies were already struggling to create durable differentiation before LLMs entered the picture. The models simply expose the gap between what a brand claims and what the market has actually absorbed.
The brands that show up well in AI-generated responses tend to share a few characteristics. They have a clear category association. They have a consistent point of view that has been expressed repeatedly across many sources over time. They have earned external validation, whether through press, awards, customer advocacy, or analyst coverage. And they have avoided the temptation to reposition every two years when a new CMO arrives.
The Signal Sources That Actually Matter
When I was building the agency’s SEO practice, one of the things we kept coming back to was the difference between signals that you control and signals that you earn. Earned signals, particularly links from authoritative sources and consistent mention in high-quality editorial, were always more durable and more commercially valuable than anything we could manufacture. The same logic applies to LLM brand representation, with some additional nuance.
The signal sources that carry the most weight in shaping how a model understands your brand include:
Editorial and press coverage. Articles in publications with genuine editorial standards, particularly those that describe what your brand does in specific terms, what problem it solves, and how it differs from alternatives. A profile piece in a respected trade publication is worth considerably more than a press release republished across wire services.
Review platforms and user-generated content. The aggregate of what customers say about you across G2, Trustpilot, Google Reviews, and similar platforms forms a significant part of the model’s understanding of your service quality, reliability, and customer experience. Volume matters, but so does the specificity of the language used. Generic five-star reviews contribute less than detailed accounts of what the experience was actually like.
Forum and community discussion. Reddit, Quora, industry forums, and niche communities are heavily represented in training data. How your brand is discussed in these contexts, whether as a recommendation, a cautionary tale, or a point of comparison, shapes the model’s understanding in ways that are difficult to influence directly.
Analyst and research coverage. For B2B brands in particular, inclusion in analyst reports, market maps, and category definitions carries significant weight. Being named in a Gartner Magic Quadrant or a Forrester Wave is not just a sales tool. It is a signal that influences how models categorise and describe you.
Consistent brand voice across third-party contexts. Consistent brand voice has always been a marker of brand maturity. In an LLM context, consistency across sources helps the model build a coherent picture. Brands that sound different in every context, different on LinkedIn than in press coverage, different in customer reviews than in analyst commentary, produce a fragmented signal that models struggle to synthesise into anything useful.
How to Audit What a Model Currently Thinks of You
This is more straightforward than most teams assume, and it should be done before any strategy is built around improving your LLM presence. The method is simple: ask the model directly, across multiple prompts and multiple models.
Start with category-level prompts. “What are the leading providers of [category]?” “Which brands are known for [specific attribute]?” “How would you describe [brand name]?” Then move to comparative prompts. “How does [brand name] compare to [competitor]?” “What are the strengths and weaknesses of [brand name]?” These prompts reveal not just whether your brand appears, but how it is characterised and where it sits in the competitive frame the model has constructed.
Run the same prompts across ChatGPT, Claude, Gemini, and Perplexity. The outputs will differ because the training data and model architectures differ. Where they converge, you have a reasonably reliable picture of the signal that has accumulated. Where they diverge, you have ambiguity in your brand’s external presence that is worth investigating.
Document the language the models use. If they consistently reach for the same descriptors, those are the associations you have earned. If they default to category generics, your differentiation work has not penetrated the external record. If they get things factually wrong, you have a specific reputational gap to address through earned coverage and correction.
I would treat this audit the same way I treated competitive positioning reviews when running the agency: as a starting point for honest conversation, not a vanity exercise. The output tells you where you actually are, not where you wish you were. Brand equity in digital contexts has always been shaped by what others say, and this is simply the most recent and most visible manifestation of that principle.
What You Can Actually Influence
The honest answer is: quite a lot, but slowly. There is no shortcut here. The inputs that shape LLM brand representation are the same inputs that have always built durable brand equity, earned over time through consistent behaviour and external validation. BCG’s research on recommended brands has long pointed to the same conclusion: brands that earn word-of-mouth and third-party endorsement outperform those that rely primarily on paid visibility.
What you can do, practically:
Invest in earned media with genuine editorial standards. Not press releases. Not sponsored content. Actual editorial coverage in publications your buyers trust. This takes time and requires something genuinely worth writing about, a point of view, a data set, a case study with real numbers, a contrarian position that holds up to scrutiny.
Build a review strategy that generates specificity, not just volume. Prompt customers to describe the experience in their own words, the problem they had, how your product or service addressed it, what changed as a result. Specific language in reviews contributes more to the model’s understanding than a score.
Create content that other sources cite. Original research, proprietary data, frameworks that practitioners find useful. When your content becomes a reference point that other writers and publishers cite, you are building exactly the kind of third-party signal that carries weight in training data.
Maintain positioning consistency over time. This is where most brands fail. Every repositioning exercise, every new CMO’s instinct to refresh the messaging, every campaign that chases a trend rather than reinforcing a consistent identity, fragments the signal. The brands that show up clearly in LLM outputs are almost always the ones that have said the same thing, in the same terms, for long enough that it has accumulated into a recognisable pattern.
Engage in the communities where your category is discussed. Not with promotional content, but with genuine contribution. Answering questions, sharing expertise, being present in the conversations your buyers are having. Community presence generates the kind of third-party mention that models pick up on.
One thing worth flagging: some teams are experimenting with tactics specifically designed to influence LLM outputs, seeding content at scale, optimising for model-specific patterns. I would be cautious about this. Models are updated, training data is curated, and tactics that look like manipulation tend to have a short half-life. The more durable approach is to build the kind of brand presence that would show up well regardless of how the model works, because it reflects genuine external recognition.
The Competitive Dimension
Here is something that does not get enough attention: how a model describes your brand is often shaped by how it describes your competitors. LLMs are trained on comparative content, reviews that compare options, articles that rank alternatives, forum threads where buyers weigh up choices. The competitive frame the model has built for your category influences where you sit within it.
This means your LLM presence is partly a function of your competitors’ positioning. If a competitor has built a strong, specific, externally validated reputation for a particular attribute, the model will tend to assign that attribute to them and look elsewhere to characterise you. If no one in your category has earned a clear association with a specific benefit, there is an opportunity to claim it through consistent, externally validated positioning.
When I was growing the agency’s European hub, one of the things that worked was being very specific about what we were: a performance marketing operation with genuine multilingual capability, not a generalist agency that happened to have some international clients. That specificity showed up in how clients described us, in the briefs we won, and eventually in how the network positioned us internally. The same principle applies here. Specificity in positioning, consistently expressed and externally validated, is what gets you a clear signal rather than a generic one.
A comprehensive brand strategy has always required clarity about competitive differentiation. LLMs have simply added a new and highly visible context in which that differentiation either shows up or does not.
The Longer-Term Implication for Brand Investment
There is a broader point here that connects to a debate I have watched play out across the industry for years: the tension between brand investment and performance marketing. Performance marketing is measurable, attributable, and fast. Brand investment is slower, harder to attribute, and requires patience that quarterly reporting cycles rarely allow.
LLMs have changed the calculus in one important respect. The outputs that brand investment produces, earned reputation, consistent external positioning, third-party validation, are now directly visible in the AI layer that sits between your brand and your buyers. A buyer who asks an LLM for recommendations in your category is receiving an output that is almost entirely a function of your brand investment, not your performance marketing spend. Your paid search budget does not appear in that answer. Your earned reputation does.
This does not make performance marketing less important. It does make brand investment more clearly valuable in a way that was previously harder to articulate to sceptical CFOs. The question “what does brand investment actually buy us?” now has a more concrete answer: it buys you presence in the AI layer, which is becoming a primary discovery channel for a growing proportion of buyers.
BCG’s work on brand and go-to-market alignment has consistently shown that brands with strong internal and external coherence outperform those where positioning is fragmented across functions. The LLM context makes that coherence requirement more urgent, not less.
Brand strategy is not a one-time exercise. If you want to think more systematically about how positioning, archetypes, and long-term brand building connect to commercial outcomes, the Brand Positioning and Archetypes hub is a useful place to work through the frameworks that sit underneath all of this.
About the Author
Keith Lacy is a marketing strategist and former agency CEO with 20+ years of experience across agency leadership, performance marketing, and commercial strategy. He writes The Marketing Juice to cut through the noise and share what works.
