There are some actors who are unforgettable in a role, and from then on, it's the only type of part you'll ever see them in. It's a shame, really - these actors have range. But casting simply stopped seeing it. They've been typecast.
Companies get typecast too. Your positioning, everything you say about what you are and who you're for, is your audition. Every time a buyer asks AI about your market, AI reads that audition and decides whether you make it into the answer, and if you do, what the buyer hears about you.
Positioning complex B2B products is the part of marketing I've loved most, and I've spent my whole career doing it. So when I realized that AI is now shaping how buyers understand whole markets, I wanted to know what it was doing to the positioning work underneath. I saw that AI was telling buyers something recognizable about companies, but it wasn't necessarily what the company would have chosen to say. The analysis that follows is me pulling at that thread.
AI doesn't relay what you say about your company, even in branded queries when you are quite literally the subject of the buyer's question. It's explaining the category to the buyer and using you as one piece of that explanation. It keeps what helps it tell that story. The rest, nearly 80 percent, gets cut.
Then it builds what's left into a role: a short description of what you are and who you're for. That description is the company the buyer meets. It's narrow by design. AI doesn't need all of you. It needs you to play one part in the story it's telling.
Take Freshservice. It started as an IT help desk tool, and it has been building outward: enterprise service management for HR, facilities, finance, not just IT. Freshservice positions itself as the modern alternative to ServiceNow for companies that want enterprise-grade service management without the enterprise complexity. A buyer asks AI when to switch from a legacy help desk to a modern service management platform. Here is part of the answer:
The buyer didn’t ask AI to compare these companies. AI volunteered the sorting because that is how it explains a category.
Look at the label Freshservice was assigned: mid-market value. It is fair. But it is not the company Freshservice is building. Freshservice is pushing into enterprise service management, displacing incumbent platforms, landing its first million-dollar deals. None of that is in the label. What is in the label is a price ceiling, and a frame that tells the buyer which competitors to weigh Freshservice against. The buyer’s mind is already shaped before a sales rep ever gets in the room. Every company whose AI label is narrower than the company it is building toward pays a version of this cost.
What makes this cost hard to manage is where it sits. It lands upstream, before your go-to-market gets a turn, so the effort you spend later goes to undoing a frame rather than setting one. It's almost invisible to you, and it hardens over time, because the longer a frame goes unchallenged, the more the rest of the web writes to match it.
Being seen and being understood are different problems. Most of the work companies do on AI today is about being seen: making sure AI mentions you at all.2 What buyers actually hear about you once you do get mentioned is what I set out to explore.
Key findings
AI filters your positioning.
Take Snyk. It is an application security company. Here is how it describes itself. The blue text is what reaches the buyer. Everything else is cut.
There is a pattern here, true not only for Snyk but for every company in this study. Only claims that explain what the product does, how it is built, where it fits, have a chance of making it through. Claims that signal trust, analyst rankings, customer counts, aspirational claims, are consistently cut. Every single one.
It took me a while to see what these claims have in common, and then it was obvious: they all exist to help a human decide. They build confidence, create aspiration, give a person something to feel. But AI doesn’t need to feel confident. AI isn’t scared of getting fired for a bad recommendation. It has no gut to validate. So it drops everything designed to move a human and keeps only what helps it explain the category.
But not all claims that explain the category reach the buyer. Some of it depends on the query: where the buyer is in the buying cycle, whether they ask about the category in general or name you directly. Late-stage questions, the kind a procurement team or a CFO would ask, let very little through regardless. Besides that, the strongest predictor I found was how clear the company’s own positioning is: how focused and single-minded.3 That is good news, because it is something you can change.
A machine reads your positioning differently than a human. Companies have been describing themselves for human buyers, and doing it well. Now a machine stands between that work and the buyer, and it keeps only what it needs.
AI does not list your claims back to the buyer. It distills them into a label, a short description that captures what it thinks differentiates you from competitors in your category. That label draws on your positioning, but it is not a copy of it. It is AI’s interpretation of what makes you different.
Each company gets the trait that most clearly differentiates it from competitors. Sometimes that is where you started. Sometimes it is your market tier. Sometimes it is one product capability out of dozens you offer.
For a given company, AI tends to land on one label and repeat it, but not uniformly. The typical company gets its most common label about two-thirds of the time.4 The remaining third is a mix of other descriptions.
Consider fleet management.
Motive started in trucking compliance, selling electronic logging devices (ELDs) that truckers are required by law to carry, then expanded into a full fleet platform with dashcams, GPS tracking, and spend management.
About 40% of the time, AI calls Motive the ELD compliance company, a label drawn from where it started. Another 30%, the AI safety and dashcam platform, drawn from a different capability. The rest is a mix of other descriptions. Two competing labels, neither dominant.
Compare that to Gong, which gets “conversation intelligence” consistently, or Freshservice, which gets “mid-market alternative” consistently. The label is stickier for some companies than others.
That consistency runs against what you would expect from AI. Recent research has found AI answers to be unstable: the shortlist of vendors, the order they appear in, and the tone all shift, even when the same category question is asked twice.5 And yet the label a company is given, when it does appear, stays largely consistent: across models, buyer personas, buying stages, and prompt complexity.6
For most companies, more than 40% of what AI says in unbranded queries is generic: it could just as easily describe a competitor.7 Visibility doesn’t protect you from this. There is little relationship between how often AI brings a company up and how differentiated it makes that company sound.
The industry itself explains about two-thirds of the difference between companies in this chart. Some markets are structurally differentiated, with natural lanes AI drops each vendor into: in contract management and construction, around 70% of what AI says about the typical company is differentiated. Others are commoditized: in OT security, AI describes every vendor in much the same terms, and less than a third of what it says is differentiated. Within an industry, though, companies can land far apart. In account-based marketing, RollWorks is described in differentiated terms about twice as often as 6sense and Demandbase. Fleet management and medical imaging are split the same way.
The takeaway is that driving your AI visibility is not enough. A generic description gives the buyer no reason to choose you over a competitor, and little reason to remember you at all.
Three things determine how AI represents your company. They are independent of each other, driven by different inputs, and fixed by different kinds of work.
| The lever | What sets it | Where your effort goes |
|---|---|---|
| Whether AI mentions you at all | How much gets written about you across the web | Visibility and coverage, which you are probably already doing |
| How much of your story comes through | How clear your positioning is, and how your claims are framed | Sharpening what you say so more of it reaches the buyer |
| How AI describes you | Your market sets the baseline, your positioning choices decide where you land within it | Claiming a label you can own, and making claims only you can make |
This is driven by third-party coverage. You are almost certainly already working on this, and there are good solutions to help you do it.
Two things determine how much of your positioning survives the filter. (How the buyer frames the question matters too, but you cannot control that.)
The first is positioning clarity: how focused, coherent, and single-minded your positioning is as a whole. AI cannot synthesize a scattered positioning into a coherent story. In treasury management, Kyriba gets nearly four times as much of its actual story through as FIS. Kyriba positions around one thing: cloud treasury and liquidity management for CFOs. FIS positions around several at once and lets AI decide which one matters. Across industries, clarity was a stronger predictor of how much positioning came through than company size, content volume, or third-party coverage.
The second is what your claims are built to do. A company with perfectly clear positioning can still lose half its story if the claims are framed as trust signals rather than structural descriptions. Effective positioning claims for AI focus on explaining what the product is and how it works.
Clarity is not the same as differentiation. You can be perfectly clear about a position that does not set you apart. But clarity gives AI something coherent to work with, so more of what you say reaches the buyer.
When positioning works, AI gives you a distinctive label and sticks to it. Snyk gets “developer-first security”, repeated consistently across every AI model, and AI describes it in more differentiated terms than any other application security company in the study.
But positioning can fail in two ways: AI gives you a label smaller than you are, or it describes you in terms that could fit anyone in your category. Different problems, different fixes:
| The wrong label | A generic description | |
|---|---|---|
| Work within your category | Claim a different label. You are in the right category, but the label AI gave you is smaller than you are. You make the case for a different one. | Sharpen the claims you make. Give AI something to say about you that it can’t say about a competitor. |
| Change your category | Play in a different category. Choose your competitive frame instead of inheriting it. | Redefine the category itself. Change how the category is understood, so the frame is one you set. |
Where to focus depends on your market. If it is structurally differentiated, AI probably already describes you in differentiated terms, so spend your effort on coverage and how your story comes through. If it is split, your positioning choices decide the outcome, and that is where effort pays off most. If it is commoditized, changing how AI describes you is much harder. It’s a strategic bet, not an incremental one.
The implicit safety that not every buyer reads the same version of you is becoming a thing of the past. You can still address an objection on the fly, reframe a capability during a demo, respond to a concern the buyer hasn’t articulated yet. But the buyer is increasingly walking in with a view already formed by AI, and your live effort goes toward undoing that view instead of shaping a new one.
Positioning has always been hard to measure. Brand sentiment, win/loss analysis, focus groups: none of them tell you whether your intended differentiation is actually landing with the buyer. Now AI hands you that measurement directly. Structured, repeatable, comparable. You can literally read how your buyer understands you. The typecast actor never finds out what they were reduced to. You can.
And that measurement is not just a diagnostic. The labels AI assigns when a buyer asks about your market are becoming the context in which purchasing decisions get made, by humans today, and by buying agents acting on their behalf tomorrow. I tested a subset of the same companies with prompts ranging from simple category questions to structured procurement simulations with weighted scoring criteria. The labels did not change.8 The window of opportunity for influencing this is open right now.
When AI describes a category the same way every time, in words that could fit any vendor, buyer perception of differentiation compresses. What does that mean for how you compete, price, and evolve?
If you've seen similar patterns in your category, or the opposite, I'd like to hear about it. I'm interested in examples that corroborate these findings, examples that contradict them, and especially in real buying decisions: what a buyer asked AI, what it answered, and whether it changed the shortlist.
I also run this analysis for individual companies. If AI’s version of your company doesn’t match the company you’re building, that is what this work is for: maya.poleg@gmail.com
1 The percentages in this chart come from checking each positioning claim against each AI model’s answer to a relevant buyer query. Every query was run three times per model, and I took the majority answer. To keep industries with more companies from skewing the averages, each of the 15 industries is weighted equally. Scoring was done by AI (Claude Opus), with a second model (Gemini 2.5 Pro) used to cross-check consistency. The exact percentages are approximate: a second scorer would land within about 10 to 15 points on any single rate, but the patterns, which types of claims get through and which do not, hold regardless of who scores. ↑
2 Two things decide whether AI names you. You need independent coverage, other people writing about you, so you exist in what AI reads. And you need to be filed in the category the buyer asked about. How much you publish showed no relationship to how much of your story gets through. Once you are named, whether your positioning comes through is a separate question, and it turns on what you say: how you frame yourself, whether you fit the category, whether your claims are clear. Not your content volume. Not your citation count. Coverage gets you into the room. It does not get your story told. ↑
3 Clarity is a company-level score, not a per-claim measure. For each company, I scored positioning clarity on a 1-to-5 scale from their website and marketing materials, blind to the AI output. A 5 means the company’s positioning is focused, coherent, and single-minded. A 1 means the positioning is scattered or vague enough that a reader would struggle to say what the company does. I then tested clarity against company size, claim architecture, content volume, and third-party coverage to see which best predicted how much of a company’s positioning survived in AI answers. Clarity outpredicted all four in 13 of 14 industries (one industry had too few companies to test). Separately, a company’s industry explains about two-thirds of the differences between companies in how generic AI’s description of them is. The rest varies by company within the industry. This is consistent with independent research finding that how AI treats a brand is largely set by the brand’s position in its market: category leaders are recommended consistently regardless of who is asking, while mid-market brands are the ones whose treatment moves (Jack, Lehman, Maloney and Xu, arXiv:2605.30207, 2026). ↑
4 The label-concentration figure is based on 70 of the 75 companies. Five were excluded due to insufficient data. For each remaining company, I read every AI description and classified it into a label. Concentration is the share of descriptions that land on the company’s single most common label. The average concentration across the 70 companies is 64%, reported here as “roughly two-thirds.” ↑
5 Identical prompts rarely return identical answers. In one study, 12 prompts run nearly 3,000 times across ChatGPT, Claude and Google returned an identical list in under 1 run in 100, and the same list in the same order in about 1 in 1,000 (Fishkin and O’Donnell, SparkToro and Gumshoe, January 2026). In another, the set of brands mentioned overlapped by only 45 to 59 percent between consecutive days across four AI engines over 45 days (Schulte, Bleeker and Kaufmann, “Don’t Measure Once”, arXiv:2604.07585, 2026). ↑
6 I tested the two-thirds figure across several dimensions to make sure it holds broadly, not just on average.
| Dimension | % of companies where the label holds |
|---|---|
| By buyer persona (CFO, engineer, functional leader) | 72% |
| By buying stage (awareness, evaluation, decision) | 75% |
| Branded vs. unbranded query | 75% |
| Across AI models | 72% |
7 Unbranded queries only. I broke each company’s description into individual statements and checked each one against what the same answer said about its competitors. A statement is generic if it only describes the category (“unifies customer data”), says nothing concrete (“a strong option”), or repeats what AI said about a competitor. Everything else is differentiated. “How often AI brings the company up” is the share of unbranded answers in the company’s industry that describe it, not just name it. Scoring was done by AI following a written rulebook. Two independent scorers landed within 10 points of each other on 78% of company scores. ↑
8 Which vendors appear in any given AI answer varies from query to query (see note 5). What each vendor is called when it does appear is what stays largely fixed: the labels in this study held across models, buyer personas, buying stages, and prompt complexity (see note 6). ↑