Questions and answers

The basics

What is generative engine optimization?

Generative engine optimization, or GEO, is the practice of making a company appear in the answers that AI assistants give when someone asks for a recommendation. When a buyer asks ChatGPT for the best software in a category, the model returns a handful of names. GEO is the work of being one of them.

What is the difference between GEO and SEO?

SEO competes for a position in a list of links. GEO competes to be named inside a written answer. The mechanics differ in one important way. A search engine matches keywords and returns pages. An answer engine reads pages and extracts entities, then decides which names belong in the response. That means a page optimized to rank can be read by a model and still produce no mention of the company that wrote it. We measure that gap directly.

Does GEO replace SEO?

No. Ranking well still helps, because pages that rank are more likely to be retrieved and read. The two overlap but they are not the same job, and doing one well does not guarantee the other. In our study of 85 B2B software companies, only 47 percent of answers that named a company also cited that company's own website. Most of the time a model describes a company using someone else's page.

Why does being named actually matter?

Because a shortlist is where a purchase begins. If a buyer asks an assistant which vendors to consider and a company is not in the answer, that company never enters the evaluation. It is not outranked. It is absent from the decision entirely.

Is this a real channel yet, or is it early?

It is early, and we say so. Buyers are clearly using assistants for software research, but nobody can hand you a clean attribution model for it today. What we can do is measure presence, which is the necessary condition for anything downstream. We would rather show you a measurement than a projection.

How we measure

What exactly do you measure?

Two things per answer. Whether the model named your company, and whether it cited your domain as a source. Those are separate outcomes and they diverge more than most people expect. A company can have its own page pulled into an answer and still not be recommended in that same answer.

Which AI engines do you check?

Four. Claude, ChatGPT, Perplexity and Google AI Overviews. We run the same question to all four so the results are comparable, and we report each engine separately rather than blending them into one number. Engines disagree with each other more than they agree.

How do you choose the questions?

The questions are generated from your category before anyone on our team reads your website. That order matters. Questions written after reading a site tend to be shaped, without meaning to be, toward what that site already says well. Ten questions per run, covering category questions, alternatives to named competitors, direct head-to-head comparisons, use case questions and pricing.

How accurate is this? Do the answers change between runs?

They change, and we tell you by how much. AI answers vary run to run. That is why we report per-engine splits instead of a single blended score, and why on larger engagements we run the same category more than once with independently generated question sets. Numbers that hold across runs are a position. Numbers that move are a reading. We label which is which.

What does it mean if AI cites my site but does not name my company?

It means the model found your page useful enough to source a fact from, and then recommended somebody else in the same answer. Across 85 companies we measured, 29.4 percent were cited more often than they were named, and nine were named in exactly zero answers while their own domain still appeared in the citation list. Usually it is a framing problem rather than a content quality problem.

Do you measure sentiment or just presence?

Presence. Being named is not the same as being recommended well, and we do not claim to measure the difference. If a model names you dismissively, our score counts it the same as a glowing mention. That is a real limitation of the method and we would rather state it than hide it.

The two free things

What is the difference between the instant check and the full audit?

The instant check is the widget on our homepage. Five questions through two engines, results by email in about ten minutes, no conversation required. The full audit is the larger version. Ten questions through four engines, every citation opened and classified by hand, and a two page written findings document. Both are free. Neither requires a call.

Why do you not require a call first?

Because a call before findings is a sales meeting, and you have no way to judge whether we are any good until you see our work. Sending the findings first inverts that. If the document is useful you will know without us telling you, and if it is not, you have lost nothing but the time it took to read two pages.

What is in the two page findings document?

How many of forty answers named you and how many cited your domain, broken out by engine. Which brands were named instead and how often. All ten questions with your result on each. What the engines actually read to write those answers, with the publisher of every top source identified. Then a diagnosis and the three things we would fix first, in order.

What is the catch?

There is none, and the honest reason is that this is how a firm with no reputation earns one. We would rather spend the cost of an audit than the cost of a sales team. If the findings are useful and you want the work done, we will quote you. If not, keep the document.

Will you spam me afterwards?

No. If you do not reply, we send one follow up and then stop.

Working together

How much does it cost?

Three thousand dollars a month with a three month minimum. There is no setup fee and no long term contract beyond those three months. We publish the price rather than qualifying you into a quote.

What happens in each month?

Month one is entity foundation. We fix how the engines identify your company: name consistency, the ownership and category record on the third party sources they read, and the structured data on your own site. Month two is the answer layer. We publish the pages that answer the buying questions your category has left unowned, including honest comparisons naming your competitors. Month three is source presence, extending into the third party pages and video and transcript surfaces the engines cite. From month four it is maintenance and measurement, roughly fourteen hours a month.

How long before anything changes?

Entity corrections can move within weeks, because they are data problems on pages the engines already read. Answer layer work is slower and depends on those pages being retrieved and re-read, which is measured in months rather than weeks. Anyone promising you a fixed timeline for being named by a language model is guessing.

How many clients do you take?

Few. One person doing this properly can hold about six engagements and no more than two in build phase at once. We would rather move the price than the volume.

What can you not promise?

We cannot promise a model will name you. Nobody controls that, and any firm that says otherwise is selling certainty they do not have. We can promise measurement you can check, work on the specific gaps the measurement found, and an honest report when something did not move.

Who is this not for?

Companies with no clear category, so there is nothing to be named in. Companies whose competitors are also absent, because then there is no gap to close and no story to tell. Teams that already have someone doing this in house. And anyone who wants a dashboard rather than the work, since there are good tools for that and we are not one of them.

About us

Who is behind Broadcastwell?

A small distributed team, with people in Bloomington, Chicago, San Francisco and New York. The company is registered in Indiana. We are new, which is why everything we claim is published and checkable rather than asserted.

Is your research public?

Yes, all of it. The 2026 State of GEO covers 85 B2B software companies, 860 scored answers, 60 categories and 5,160 citations, and it is published with a DOI and open datasets anyone can download. Volume II, The Absence Ladder, classifies 616 answers where a brand was absent and asks what kind of question it lost. Both are on this site under Research.

Do you have case studies?

Not yet, and we will not invent them. What we have instead is the published dataset, the method, and the fact that you can run the same measurement on us. Ask for the raw scoring behind any finding we send you and we will hand over the whole file.

How visible is Broadcastwell in AI answers?

Less than we would like, and we publish that too. Our own category audit is on the Our visibility page, including the competitors the engines name ahead of us. We think a firm selling visibility should show its own numbers before it shows yours.

Practical

What do you need from me to run an audit?

A work email, your website, and the category you compete in as a buyer would say it. Nothing else, and no access to any of your systems.

Do you need access to my analytics or CMS?

Not for the audit. Everything we measure is public, which is the point. We are looking at what buyers see, not at your internal data.

How is this different from an AI visibility tracking tool?

A tool gives you a dashboard and leaves the work to you. We do the measurement and then do the work the measurement points to. If a dashboard is what you want, buy a dashboard. It will be cheaper than us.

Can I just do this myself?

Yes, and some of it you should. The method is published, the questions are the ones your buyers already ask, and nothing about running them is secret. What takes the time is opening every citation, working out which pages the engines actually trust in your category, and then writing the material that earns a place there.

See your own numbers first

Five buyer questions through two engines, results by email in about ten minutes. No call required.