There is a sentence about your company you have never read.
Ask ChatGPT, Claude, Gemini or Perplexity about your market and it will produce one line about you, or leave you out. That line is now the first thing most buyers learn about your company — and no one inside it has seen it. ModelVox is a positioning firm for that sentence.
Complimentary first call. You leave with the sentence, and the sources it came from.
You are being described more than you are being read.
For twenty years the path from a question to a company ran through a search box, and the company controlled the landing page at the end of it. That path now ends earlier. A model reads the market on the buyer's behalf, forms a view, and hands over a summary — a sentence or two per company, a shortlist, a recommendation.
Which means the most consequential piece of writing about your company is no longer written by your company. It is assembled, per buyer, per question, from whatever a model can find and verify. If you have not made anything specific easy to verify, it writes the sentence that is true of your entire category. Accurate. Unmemorable. Fatal.
A model does not quote your homepage. It produces its own description of you from what it can corroborate. When nothing specific is corroborated, it falls back on what is true of your category — the default sentence. You cannot write that sentence. You can make a better one the cheapest thing for a model to say.
Most of what is sold as AI-search optimisation is aimed at the wrong problem.
llms.txt files that received no requests at all in Ahrefs' study of 137,210 domains. The most-recommended tactic of the last two years is mostly unread.Sources, in order: Vercel, “The rise of the AI crawler”; Ahrefs, llms.txt study; Perplexity crawler documentation. We keep a maintained version of this on the crawler policy page.
The default sentence
Every model, asked about a company it knows little about, has a cheap option available: describe the category and attach the company's name to it. "A leading provider of AI-powered solutions for enterprise." Nobody wrote it. Nobody approved it. Nothing in it is false.
It is also the most expensive sentence in your business, because it is the one a buyer reads instead of your positioning — and because it is indistinguishable from the sentence your competitors get. Two identical descriptions collapse into one decision: price.
The work is not to argue with the model. It is to make the specific claim easier for the model to reach than the generic one. That is an evidence problem before it is a marketing problem, and it is the whole of what we do.
One firm, one discipline, two questions worth paying for.
Be the company the model names.
For companies that will be chosen or skipped on the strength of how they are described. We decide what should be said about you and build the evidence that makes a machine say it.
Make one workflow provably worth it.
For established businesses where a repeating workflow is quietly losing money. We establish the baseline, implement inside the systems you already run, measure the difference in 30 days, and then keep it working.
Different buyers, different questions, same method: establish what is true before deciding what to say, and refuse the work when the evidence will not support the claim. See all six engagements →
Four moves. In this order, because the order is the method.
Read
Collect what the systems actually say about you now — across ChatGPT, Claude, Gemini, Perplexity and AI Overviews — and trace each description back to the sources that produced it. Almost no company has read this.
Choose
Decide the sentence that should replace it: the one claim that is true, specific, defensible, and not equally true of the two companies you are confused with. This is the positioning decision, and it is the hard one.
Prove
Make the claim verifiable. Sometimes that means publishing evidence. Sometimes it means changing the company until the claim is true. We will tell you which one you are facing.
Spread
Get the claim corroborated where models already look — independent sources, not your own site. This is the part everyone skips, and the part that moves the answer.
Five expensive mistakes, in rough order of how often we see them.
Treating it as a content problem
Publishing more rarely makes a company more distinct. It usually makes it sound more like everyone else, faster — and gives a model more generic material to summarise. Volume is the one lever that can actively make the sentence worse.
Treating it as a technical problem
Schema, an llms.txt file, a tidy sitemap. All worth doing, none of it decisive. These make you readable. They do not make you worth mentioning. A perfectly marked-up company with nothing specific to say gets the default sentence in structured form.
Assuming the model has read your website
It has read some of it, at best, and weighted it lightly, because you are not a neutral source about yourself. What a model treats as evidence is what independent sources say. Your site sets the claim; other people settle it.
Building the site in the browser
Client-side rendering is the single most common cause of a company being invisible to AI retrieval. Google renders JavaScript. Most AI crawlers do not. Companies with excellent websites are being read as blank pages, and nothing in their analytics says so.
Claiming something not yet true
Corroboration is the mechanism, which means a claim your customers and the press will not repeat cannot be made to stick by writing it more confidently. If the claim is ahead of the company, the honest move is to say so and fix the order.
Two kinds of leader. One shared problem: being understood by something that never asks a follow-up question.
AI-native companies
Founders and executives at AI infrastructure, agent, robotics, healthcare, cybersecurity and developer-tools companies, in categories where every company describes itself in the same eight words. The technical work is excellent and the description is interchangeable. That gap is the entire problem.
- Raising, and finding that the deck lands and the category does not
- Confused with two named competitors in every buyer conversation
- Losing deals to companies with a weaker product and a clearer claim
- Absent from AI answers where competitors appear by name
Established businesses
Leaders in professional services, healthcare, legal, financial services, manufacturing and field services who want AI to change one number, not to run a science project. Usually there is a workflow quietly losing money and nobody internally who owns it.
- Enquiries answered too late to win
- Quotes sent and never followed up on a consistent cadence
- The same information re-typed between systems that should talk
- A suspicion that opportunities leak, with no instrument that shows where
What we have not done.
Everyone in this market has a case study. A great many of them are decorative. Since the argument we make to clients is that corroborated claims beat confident ones, it would be incoherent to list proof we cannot substantiate. So here is the disclosure, and it stays on the site as it changes.
As of August 2026
- We have published no client case studies. The work exists; the write-ups are with clients for approval. A case study a client has not read and approved is not evidence — it is marketing with their name on it — so nothing appears here until it clears that bar.
- We have no awards, certifications, analyst placements or partner badges. We have not applied for any.
- We do not have proprietary access to any model provider's ranking, and neither does anyone else selling this service.
- Our own AI visibility is a work in progress. This site was rebuilt in August 2026 against the method we sell; the corroboration layer takes months, and we are early in it.
What we do have is a method we will walk you through before you pay for anything, a first call that produces a finding you keep, and the willingness to tell you the work is not worth doing. More on how we work →
The questions buyers ask a model before they ask a firm — answered plainly.
Direct answers, before the first call.
If the answer to one of these disqualifies us, that is a good outcome for both of us and a cheap one.
When the question turns to reliability, we don't answer it ourselves.
ModelVox is accountable for how a market and its machines understand you. Whether your production agents actually work, and whether the code behind them would survive diligence, are different questions with different methods — and an independent finding is worth more than ours. Two separate firms: one audits the agent, one audits the code. Neither sells the fix.
A five-day, fixed-price CARA Audit™ that scores a production agent 0–100 across task completion, consistency, tool use, safety, and evaluation — run like a penetration test, not a demo. You walk away with a ranked 90-day fix roadmap. Building that fix is someone else's job.
A Technical Integrity Audit™ that goes under the hood — architecture, technical debt, AI-generated code risk — the diligence investors and boards run before they bet on what you've built. Built for the moment your codebase becomes a deal term, not an afterthought.
Find out what the models are saying.
The first call is a working session, not a pitch. We come having already read your sentence across the major systems, and we show you where it came from. You keep the finding whether or not we go further.
The first call is complimentary — and the finding is yours to keep.