Questions
Plain answers to what people actually ask us. Where there is a number in an answer, it carries where it came from and when it was measured. Where the honest answer is that it depends, the answer says what it depends on.
Get the claim onto pages you do not own, and make sure OAI-SearchBot can read yours. ChatGPT gets its search results from Bing, so a page Bing has not indexed cannot reach it. Then it reaches for the claim that many independent sites repeat. One page on your own site saying you are the best is one source. The same fact on Wikipedia, Wikidata, a trade publication and a review platform is a consensus, and consensus is what it quotes.
Usually because there is not enough on the open web that a machine can verify about you. An engine answering a category question builds the answer from everything it can check, and it picks the brand it can resolve: the one it can name, source and repeat without hedging. If the only material about you is on your own site, and most of that site is code rather than readable text, there is very little for it to work with.
Because it found more about them that it could check. An engine builds a recommendation from the sources it can reach, and it names the brand it can name without hedging. Your competitor is usually not doing anything clever. They are present in the places the engine reads, and you are present mainly on your own site. The useful move is to find out which sources the engine pulled them from and whether you appear in any of those sources at all. One run is also not a verdict: ask again tomorrow and the answer can move, which is why every reading Stage30 publishes carries its date and time.
You can change what it has to work with, which changes the answer. You cannot type the answer yourself. The practical work is making the true things about you verifiable in more than one place, structured so a machine does not have to infer them, and answering your buyers’ real questions in public where a crawler can reach the answer.
It depends which engine, and the spread is wide. Perplexity crawls your pages itself, so work on your own site can surface quickly. ChatGPT waits on Bing. Gemini is the slowest of the four, because it starts from the opinion Google has built about your domain over years. Anyone promising all four on the same schedule has not looked at how they differ. Set a baseline before you start, or you will have no way to say what the work did.
AI search optimization is the work of making a brand the name an AI engine returns when a buyer asks a question in its category. The unit of the work is a claim about the brand, repeated across enough independent and machine readable sources that an engine can state it without hedging. In practice that is three things: structured entity data on your own site, the same facts verifiable on sources you do not control, and answers to your buyers’ real questions published where a crawler can reach them. Stage30 measures the result as a Resolution Score out of 100, and the free report shows which of the three is holding you back.
AI search visibility is whether an AI engine can find, verify and name your brand when it answers a question in your category. It has a floor and a ceiling. The floor is whether the engine can read you at all, which depends on crawler access and on whether your pages carry text a machine can quote. The ceiling is whether the engine names you first when several brands qualify. Most brands that check assume they have a ceiling problem and find they have a floor problem.
Ask the engines the question your buyer would ask, record the whole answer, and count whose name comes back. That is the measurement, and everything else is inference. A Stage30 Brand Resolution Report puts your buyer’s question to ChatGPT, Perplexity and Claude live, quotes each answer word for word, lists the sources each engine pulled from, and scores the result out of 100. Run it twice on different days before you trust a single number, because the engines move. Every reading Stage30 publishes carries the date and time it was taken, for that reason.
SEO and AI search SEO share a toolkit and answer different questions. SEO earns a position on a list of links, and the buyer still chooses from the list. AI search SEO decides which single brand the engine names when it writes the answer, so the list never reaches the buyer. The overlap is real: crawlability, clean markup, fast pages and credible links matter to both. The divergence is that an engine reads the whole web’s account of you rather than your site’s account of itself, so work that never touches your own pages, such as a corrected Wikidata entry or a trade citation, can move the answer further than a homepage rewrite.
Any brand whose buyers open with a question rather than with a brand name. The test takes a minute: type your buyer’s actual question into ChatGPT and see whose name comes back. If a competitor’s name comes back, the engine has settled your category already and you are outside the decision. If your own name comes back hedged, wrapped in ’options include’ or ’you may want to consider’, you are on the list and the answer belongs to someone else. Considered purchases, professional services and B2B feel this first, because those buyers ask the most questions before they ever type a brand name.
Change what it reads, then measure whether the answer changed. There is no correction button and no support ticket for a brand fact. An engine builds each answer in the moment from sources it can reach, so a wrong answer is usually a wrong or missing source, repeated. Find the pages the engine actually cited, correct the ones you control, get the true version onto the ones you do not, then ask the question again on a date you write down. It is a measurement problem before it is a content problem.
There is no meaningful correction channel for a brand fact. You can report safety problems, and you can control crawling through robots.txt, and neither of those edits an answer. The leverage sits in the sources, so the first job is knowing which sources the engine used. The free report names them.
Ask it your buyer’s question, more than once, and write down what came back. Asking once is an anecdote. The engines build each answer fresh, so the same question can name you on Tuesday and skip you on Wednesday. A reading is a set of runs, each one dated. Stage30 puts the question your buyer would type to three engines and quotes each answer in full, so what you are reading is the answer itself rather than a score standing in for it.
A reading, from 0 to 100, of how confidently an AI engine can pick your brand as the answer. It has two halves. One moves with how many engines named you on that run, and it changes between runs. The other comes off your own pages and holds still. That second half is the part any work you do actually moves, which is why Stage30 reports both.
Because AI keeps no standing ranking. Each answer is built in the moment, and it can land differently for another person on the same day. Stage30 has measured one site 35 times on one question and seen it read anywhere between 32 and 70. Every reading it publishes is dated to the minute for that reason, and one run should never be read as your brand.
More than once, and the vendor should tell you the number. On the site Stage30 has measured most, 35 readings of one question over two weeks ranged from 32 to 70. A single run sitting inside a spread that wide tells you very little on its own. What is worth reading is the floor: the part of the score that comes off your own pages and holds still between runs. Stage30 reports both halves for that reason, and prints the run count and the timestamp on every reading.
They asked different questions, at different times, from different places, and most do not publish how the score is built. Two numbers built that differently cannot be compared, and averaging them produces a third number that means less than either. Read a trend inside one consistent method and ignore the gap between tools. If a tool will not show you the question set, the run count and the date, the score is decoration.
Read the referral traffic you can see, and label the rest unmeasured. Most AI answers are read without a click, so the referrals in your analytics are a floor rather than a total. Anyone handing you a complete attribution model for AI answers has built it out of assumptions. The part that can be measured honestly is whether you are named, on which question, by which engine, on what date.
No, and the difference is the list. SEO competes for a position on a page of links, where the buyer still chooses. AI search returns one answer, often with no list at all, so there is no second place to be. A site can rank first on Google and go unnamed by every AI engine, and Stage30 sees it on real runs.
It matters as an input, and it is not a proxy. ChatGPT reads Bing, so a page Bing has not indexed cannot reach it, and the technical work that gets a page indexed is the same work it always was. Position is a different story. A page can rank first on Google and go unnamed by every AI engine. Keep the indexing work. Stop reading rank as a visibility score.
GEO, generative engine optimization, aims to get your content quoted inside an AI answer, as one source among several. Stage30 works on being the brand the engine names. Getting cited is useful. Being the answer is the outcome that produces a customer.
Sometimes, because each one reads a different copy of the web. ChatGPT reads Bing, Gemini reads Google, Claude reads Brave, and Perplexity reads its own. Each copy holds a different version of your brand, so one question can come back four ways. When they do agree, that agreement is worth having. There is more on this, engine by engine, on the engines page.
Block the ones that only take, and allow the ones that can send you a customer. GPTBot feeds OpenAI’s model training. OAI-SearchBot is what decides whether you can appear in a ChatGPT answer. Blocking GPTBot keeps you out of training and leaves you findable. Blocking OAI-SearchBot makes you invisible to the largest AI engine there is. The same distinction matters at Google: Google-Extended governs AI use, while Googlebot is what actually crawls, so blocking Googlebot to stay out of AI takes you out of Google Search with it.
One decides whether you can appear in a ChatGPT answer, the other feeds model training. OAI-SearchBot surfaces sites in ChatGPT’s search features. GPTBot crawls content that may be used to train OpenAI’s foundation models. ChatGPT-User is a third agent, which fetches a page when a person asks for it directly. Blocking GPTBot keeps you out of training and leaves you findable. Blocking OAI-SearchBot makes you unfindable in the largest AI engine there is, which is the expensive version of the same decision.
No. Google-Extended has no user agent of its own and never requests a page. Googlebot does the crawling, and Google-Extended is a robots.txt product token controlling whether that crawl may be used for Gemini and other Google AI products. Google states it does not affect a site’s inclusion or ranking in Google Search. The practical consequence: blocking Googlebot to stay out of AI takes you out of Google Search with it, and Google-Extended is the control that separates the two.
It is a plain text file at the root of your site that lists your important pages in language a model can read, in the way robots.txt lists rules for crawlers. It is cheap to publish and it will not rescue a site that has nothing readable on it. Publish one after your pages say something worth pointing at, not before.
It makes your facts unambiguous, which is worth doing, and no engine has published that it is a citation factor. Structured data states in code what your page states in prose: who you are, what you sell, what it costs, who wrote it. A model that would otherwise have to infer those things does not have to. Treat it as removing guesswork rather than as a lever with a promised return, and be careful with anyone selling it as the latter.
Yes. Enter a URL at stage30.com and Stage30 puts your buyer’s question to three engines, live. The report gives you that question, generated from your own page, and quotes each engine’s answer in full. It names who came back instead of you. The score is out of 100, split into the half that moves between runs and the half that comes off your own pages and holds still. Every source the engines read is linked, so you can open the page that decided it. Stage30 counts the findings on your live site and prices the fixes. No email and no card. The paid step is the $500 Resolution Audit, which opens every finding, prices each fix, puts them in order, and ends with thirty minutes on a call.
The Stage30 Resolution Audit is $500, once. You get every finding the free report counted but did not open, each one priced and ordered by what moves the answer fastest, and thirty minutes with the two strategists who wrote your report. You leave with the plan in plain language, yours to keep. Your team can run it, or we can. If we find less than $500 of work worth doing, we tell you that instead.
No. There is no placement to buy inside the answer itself. Paid units, where an engine runs them, are a separate surface and should never be reported to you as visibility. What money buys is the work: getting the true claim onto sources the engine reads, and making your own pages readable to the crawlers that decide.
Stage30 does not sell a monthly retainer. The report is free, the Resolution Audit is $500 once, and build work is quoted against the fixes your own run found. Agencies in this space commonly price monthly and the range is wide. Before you sign one, ask for the question set, the run count and the dates behind whatever number they showed you. If they cannot produce those three things, you are buying a dashboard.
Cast Iron LA, an advertising and AI consulting company in Los Angeles. The methodology is by John Otto Barbush and Luis Emilio Ramirez. Stage30 is a U.S. trademark application, serial no. 99719523, published for opposition 1 September 2026.
Because Stage30 measured the noise first. Ask the same engine the same question about the same unchanged site three times and the answers differ. On one brand in our own corpus the score read 51, 66 and 51 across three runs six minutes apart, with every page-side score identical and nothing on the site changed. So a Stage30 reading is the median of three runs rather than a single draw, and a movement is only reported when it is larger than what your own record moves by on its own. The range your score has actually occupied travels with every number, so you can see the weather rather than take our word for the forecast.
Enter your site and we put your buyer’s question to ChatGPT, Perplexity and Claude, live. The check is free and takes about a minute.
Run my free report