# AI now speaks for you. Your customer asked AI about you this morning. We check that answer against what you publish, specify what needs to change and measure every month. ## Six steps, every time. 1. We ask the four assistants. 2. The answers are kept, dated, exactly as they came back. 3. We check them against what you publish. 4. When something is wrong, we find where it came from. 5. We specify the correction, and where it has to be made. 6. We ask again the following month. ## How people choose now. In June 2026 rater8 published a survey of around 1,000 US adults. 47 percent had used an AI assistant to research a healthcare provider, up from 31 percent nine months earlier. Among people who had actively looked for a provider in the past year, AI was named as an influence by 36 percent, ahead of Google search at 34 percent and doctor recommendations at 32 percent. The report also names what the assistants get wrong. Wrong address. Wrong opening hours. Healthcare is where somebody measured this properly. The behavior is not specific to healthcare and neither are the errors. A wrong address is a wrong address whether the customer is booking a consultation, a table or a school tour. Nobody owns this answer yet. It is being written now, out of whatever is readable about you today. Source: [rater8](https://rater8.com/blog/2026-patient-choice-report/) ## What we found. We ran our own check. 40 businesses, Ho Chi Minh City, August 2026, four AI assistants, in English and Vietnamese. At least half had a verifiable factual error in what the assistants said about them, checked against each business's own published information. The two most common errors were the address and the opening hours. The same two the survey names, on the other side of the world. Some answers were not about the business at all. One returned a street address that does not exist anywhere, because it came from a template. > 123 Main Street, Anytown, USA Another was answered with a hotel in Virginia City, Nevada. Another with a restaurant in Houston. Another with a company in Manila. Ho Chi Minh City is the first market we measured. Dubai followed in August. Bangkok is next. Method: these figures describe the assistant's generated answer and do not include paid placements. Measured on Sandria, our own instrument. ## Two different problems. - **Answer accuracy.** What AI says about you, when nobody is on your site. Whether you come up at all, and whether the facts are right. - **Action integrity.** What agents can do with you, when one arrives and tries to get something done for your customer. Whether it can find your price, see what is available and finish the booking or the appointment or the application. ## Five stages, the same for everyone. This is how we work through your business, and it is the same sequence every time. - **Discovery.** Are you a candidate at all. - **Identification.** Does the agent have the right business and the right facts. - **Evaluation.** Can it get what it needs to compare: price, availability, eligibility, specifications, terms. - **Action.** Your commitment point. The one step that turns interest into commitment, whether that is a paid booking, a confirmed appointment, a submitted application or a scheduled call. - **Confirmation.** Do you answer in a way the agent can check, so the customer is told the thing actually happened. Only stage four changes from one business to the next. The other four are identical everywhere. Answer accuracy is stages one and two. Action integrity is three, four and five. ## The stage that fails. Agents rarely fail at stage four. They fail at three, on what it costs and whether it is available, because that information sits behind a form, inside an image or nowhere at all. Payment is usually not yours to fix. It belongs to your booking engine, your practice software, the marketplace you sell through. Making your prices and availability readable is yours, and it is the part that decides whether the customer ever reaches stage four. ## Where the answer comes from. An assistant does not look your business up. It assembles an answer out of whatever it can read: your site, your listings, directories, review sites, articles, old press, forum posts and other businesses with names close to yours. So when a fact is not written down anywhere a machine can read it, the answer does not come back empty. It gets filled with something else. That is how a restaurant in Houston ends up as your address. This is why the work happens at the source instead of arguing with the answer. There is no button that corrects an assistant. There is only making the fact exist, in a place it can be read, in a form it can be trusted. ## What it costs. Prices in USD. We publish them because you should be able to decide whether we are worth a conversation before you have one. ### Blueprint USD 2,500, once Your answers on four assistants today, the errors named and the list of what has to change and where. One language. Credited in full toward your first month of Presence or Authority. ### Presence USD 4,500, per month One language. 25 questions a month, across all five stages, weighted toward evaluation. We check what comes back, verify what looks wrong, then specify the correction and confirm it landed. Rechecked every month. ### Authority USD 9,500, per month Three languages, so 75 questions and 300 checks a month. Your customers do not all ask in English, and the answer is not the same in every language. Everything in Presence, and then the part that changes what AI has to work with. We find the publications and sites assistants actually quote in your category, and we pursue them for you, reported source by source. Placement depends on an editor saying yes, so it is pursued and never guaranteed. ### Implementation from USD 1,500, quoted per job, one time, never monthly The specification is included in Blueprint, Presence and Authority, so you always know exactly what has to change. Most clients hand it to their own developer. If you would rather we made the changes ourselves, we quote that as its own piece of work. ### Masterclass On request, per session For teams who want to run this in house. A working session on how assistants read your business, what they get wrong and what your people can change without us. Run on your own answers, not on general examples. ### Agentic Access On request, scoped per project An AI concierge for your business, answering your customers 24 hours a day in your languages, trained on everything you do. Also available, beside the ladder: - **Implementation**: quoted per job, from USD 1,500, one time - **Masterclass**: in-house team training, on request - **Agentic Access**: on request ### Free AI check Free, one finding, no strings Not sure where you stand? Get one verified finding about what AI says about your business, free, by email within 48 hours. One per business. ## Questions, answered. ### Why doesn't AI recommend my business? Because it either cannot find you, cannot verify you or found a competitor easier to cite. Assistants build recommendations from what they can read: your website, your business profiles, your reviews and the articles that mention you. ### What is the Blueprint and what do I get? Your answers on four assistants today, the errors named and the list of what has to change and where, for USD 2,500, credited in full toward your first month of Presence or Authority. One language. You get verified findings, a prioritized fix roadmap and a 45-minute walkthrough. ### Which languages? You choose. One at USD 4,500, three at USD 9,500. We check in the languages your customers use, not the language your website is in. The answers differ. A question asked in Arabic and the same question asked in English can come back with different facts about the same business, and the one your customer sees is the one that matters. ### Can I pay to be recommended by AI? No. There is no ad product, listing fee or fast lane that buys a recommendation in an AI answer. ## What AI says about you is your reputation now. It is written from sources you can fix. ## For agents. This site publishes a machine-readable copy of itself, kept in step with the pages a person reads. An index of the whole site is at /llms.txt. The full text of every page is at /llms-full.txt. Every content page has a markdown twin at the same path with .md on the end, linked from the head of that page. Each page also carries schema.org structured data as a single JSON-LD graph. That is the whole estate. There is no API and no MCP endpoint. - [/llms.txt](https://gkzlabs.com/llms.txt) - [/llms-full.txt](https://gkzlabs.com/llms-full.txt) - [/index.md](https://gkzlabs.com/index.md)