The counterintuitive truth about AI visibility: Most brands are trying to win elections before voters know their name.
Everyone wants to rank for queries like "best project management software" in ChatGPT. In SEO, we’d start with that keyword and build towards it with longtail queries like “best project management software for small teams.”
That’s the wrong approach with AI Search. Before we start telling the world what we can be, we need to ensure there’s a high level of clarity with who we are.
Here are some real starting points we’ve identified from our work in GEO since 2024:
- A Fortune 100 company still appears under its legacy brand name, months after a major rebrand
- A lack of intentional positioning work leads to a major brand not being considered for any specific type of customer
- A company whose employees have averaged 10+ years of tenure gets characterized as having "high turnover" because of one negative review
These inaccuracies & shortcomings in basic brand canon can’t be ignored. They’re the frame in which we start to piece the puzzle together. And it transcends the (arguably more important) need to ensure brand accuracy.
- The Fortune 100 company’s brand legacy may hold truths counter-intuitive to their current target audience.
- The major brand’s lack of positioning leaves no categories for the LLMs to select them for
- The company with allegedly high turnover is incorrectly removed from the consideration set of prompts highly relevant to their core offering.
To say it differently: LLMs are assigning more meaning to these “facts” about brands than traditional search engines. Those facts populate something akin to a thought process in the LLM, which can lead to including the brand in irrelevant responses or excluding the brand in relevant responses.
If we're already aiming at a moving target, why not make that target as high likelihood to hit as possible?
I think correcting brand inaccuracies is the best approach, not because it's easy, but because every other AI visibility effort collapses on a foundation of wrong information.
Why "Defensive" AI SEO Isn't Defensive At All
I frame AI visibility as a maturity model because the industry is grasping at straws and needs new heuristics to guide their thinking. Approaching GEO the same way we approach traditional SEO is like pouring water into a cracked vessel.
Starting with brand accuracy means fixing what's wrong about your brand first. Before you compete for category terms, you need to ensure LLMs accurately represent your basic brand facts: your location, leadership, services, and perhaps most importantly, your positioning.
There's virtually no brand that has perfect representation across LLMs of what they do, who they are, who's in charge, where they serve, who they serve, what they serve. Even established companies face systemic inaccuracies that poison every downstream effort.
The Progressive Path Most Brands Should Follow:
| Stage | Type | Example Query | Goal |
|---|---|---|---|
|
Stage 1
Pure Branded Queries
|
Defensive | "Is [Your Brand] a good agency for GEO?" | Fix how LLMs respond to queries about your brand directly |
|
Stage 2
Branded + Attribute
|
Transitional | "[Your Brand] GEO services for enterprise brands" | Optimize for queries that include your brand name plus specific offerings |
|
Stage 3
Long-Tail Non-Branded
|
Offensive | "Best GEO agencies for enterprise financial brands" | Target category terms with qualifiers that match your positioning |
|
Stage 4
Pure Non-Branded
|
Advanced | "What are good agencies for GEO?" | Compete for broad category terms without qualifiers |
This isn't a binary switch; it's a continuum requiring staged optimization.
Skip stages, and you're optimizing for a brand story that doesn't exist in the models.
Build Your Brand Canon First (Or Watch AI Tell the Wrong Story)
Before you track AI visibility or chase category rankings, you need to document what "accurate" actually looks like. Most brands can't answer this question with precision.
1. Identify 50+ Critical Brand Attributes
- Company facts: founding year, locations, size, structure
- Leadership: executives, spokespersons, founders
- Services/products: what you offer, what you don't
- Positioning: who you serve, industries, company size
- Differentiators: methodology, approach, values
- Recent changes: rebrands, acquisitions, pivots
2. Create 100+ Test Prompts (2:1 Ratio)
For each attribute, write multiple prompts that should elicit accurate information:
- Direct: "Where is [Your Brand] headquartered?"
- Indirect: "Tell me about [Your Brand]'s office locations"
- Comparative: "How does [Your Brand] compare to competitors in [category]?"
3. Test Daily and Calculate Accuracy Percentage
Run these prompts across multiple LLMs daily, not just once. These models are probabilistic. They're gonna pull in some information sometimes, they're gonna always pull it in, and then there's gonna be things that are a bit more rare.
Track this: What percentage of your brand attributes appear accurately and consistently?
This percentage accuracy score becomes your leading indicator, so track it before you track share of voice or category rankings.
4. Investigate Every Inaccuracy
For each failure:
- Check citations in LLM responses
- Return to Google SERPs for manual investigation
- Identify all web locations where the inaccuracy appears
- Map citation propagation (original source + downstream mentions)
The Skill That Separates Winners: Investigation Over Dashboards
AI Search success requires returning to investigative fundamentals that many SEOs outsourced to tools years ago.
A lot of our work has been optimized for tooling where you're spending all day in a SEMRush or Ahrefs. You gotta go straight to the source sometimes.You can't dashboard your way to brand accuracy. You have to investigate it.
Your Inaccuracy Investigation Process:
- Run prompts manually, repeatedly
Test the same prompt daily across multiple LLMs. Build pattern recognition: What appears consistently? What's rare? Don't trust one-time results (probabilistic models = variable outputs) - Check every citation
When LLMs provide sources, investigate each one. But remember: it's correlation. It's not causation. Citation presence doesn't guarantee it caused the output. - Return to Google SERPs
Search for your brand + the inaccurate information. Find where that narrative appears across the web. Map the propagation: original source + downstream mentions. - Conduct deep research combining manual and AI tools
Use AI tools to scale the search for inaccuracy sources. But verify manually, as tools miss context and nuance. Track every location where correction is needed.
This is slow work. Unglamorous work. But it's the work that needs to be done if you want the best shot at influencing AI Search KPIs.
Why Brand Inaccuracies Are Stubborn: The Training Data Problem
You might fix every mention of your old headquarters location on your website and still see LLMs cite it for months. Why?
LLMs are trained on massive historical datasets like Common Crawl, which are snapshots of the internet taken at specific points in time. Your established brand presence in these historical archives is both an advantage (you're known) and a liability (outdated information persists).
What This Means for You:
- Fixing web content is necessary but not sufficient; you're correcting the present and future, but the past still influences outputs
- Timeline expectations matter: brand accuracy improvements can take months as new training data is incorporated
- Established brands have it worse; more historical documentation means more inertia to overcome
- Consistency across time is key: the longer accurate information exists online, the more it weights future training cycles
This is why brand accuracy work needs to start NOW. The sooner you establish accurate information across the web, the sooner it begins influencing future training cycles.
With this, you're fixing today's outputs and you're shaping tomorrow's training data.
Is AI Visibility Work Right for Your Brand Now?
The threshold for shifting brand accuracy work to category-term competition (or starting at all) varies by several factors:
Start with Brand Accuracy if:
- You're an established brand with >5 years of digital presence
- You have existing brand search volume (people already search for you)
- You have the documentation to define "accurate" (you know your own brand story)
- You're in a B2B space where buying cycles are long (brand perception compounds over time)
- You have evidence of inaccuracies (run 10 test prompts; if any fail, you have work to do)
You might not be ready if:
- You're pre-product/market fit (your brand story is still evolving)
- You have zero brand search volume (pure brand accuracy work won't drive discovery)
- You lack resources to implement changes (identifying problems without fixing them creates frustration)
Move to Non-brand Category Terms when:
- You've achieved strong brand accuracy across core attributes
- You have solid citation infrastructure (press, reviews, directories listing you accurately)
- You're resource-rich enough to target 'the moving target'
- Your competitive positioning is strong (how well poised are you to lead the pack?)
- You're ready for longer-term investment (non-branded visibility builds slowly)
The Real Starting Point
My basic thesis is this: You really gotta start by making sure that your web content and your own digital assets are telling the right story first.
Fix what's broken about your brand representation. Build the canon. Establish accuracy as your leading indicator. Investigate inaccuracies like a journalist, not a dashboard jockey.
Then expand toward non-brand category dominance. I think you’ll find that changes the game from checkers to chess. Doesn’t mean it will be easy, but it does mean we can be more calculated in our approach.
If your brand foundation is cracked, everything you build on top of it will collapse. We can help you with that.
Alisa Scharf
VP, AI & Innovation
.png)
.png)