The words AI Otel does not use: why “first”, “autonomous” and “certified” stay out
AI Otel keeps a short list of words it will not use, from first to certified. Each refusal rests on a fact you can check, helping you test any vendor's claims.
Contents
Key takeaways
- AI Otel keeps a short list of words it will not use about itself, and each entry carries a one-sentence reason a reader can check.
- “First” is out because the category moves: a foreign provider announced connections for AI agents in September 2025 and a domestic one in October 2026, so the word would rest on a comparison nobody can maintain.
- “Autonomous” is out because the PMS assistant's 207 actions (June 2026) sit in three safety tiers: 107 run directly, 90 need one confirmation, 10 need two.
- “Certified” is out because no security, privacy or payment certificate exists; a readiness review, an internal policy or a self-assessment is not a certificate.
- “Trained” is out because there is no trained model yet; what exists is the architecture, the decision infrastructure a model would run inside, and a 30-case benchmark that has not been run.
AI Otel describes itself with five status words, from Live to Roadmap, and keeps a short list of words it will not use. “First” is out because others announced similar connections earlier. “Autonomous” is out because a person confirms anything sensitive. “Certified” is out because no certificate exists. “Trained” is out because there is no trained model yet.
A hotel owner opens a software brochure and meets the same three words in the first paragraph: first, autonomous, certified. The words are easy to write and hard to check. This article is the other half of the five status words AI Otel uses: the words AI Otel's language rules keep out, and the fact behind each refusal. After reading it, you will be able to put the same test to any vendor's brochure, including this one.
Why write down the words you will not use?
Because a word you refuse to use is easier to check than a promise you make. A written list turns a marketing habit into a rule an editor can enforce before publication.
Every claim has three parts: the word, which is what is asserted; the status, which is how far it is true today; and the date, which is when it was measured. Remove any one of the three and the sentence can no longer be checked by the person reading it. The status words carry the middle part: Live means operating today, which applies to OtelCiro's client work; Built means working software ahead of launch; then In development, Architected and Roadmap. Nothing in AI Otel's technology is Live.
The refused list is short, and each entry has a reason that fits in one sentence. A long list of principles nobody can apply is worse than a short one that editors actually use. The list protects the buyer as much as the writer: a vendor who cannot say what is not true yet is asking you to price a risk you cannot see.
The discipline costs something. Sentences read flatter, headlines get weaker, and some paragraphs take twice as long to write. That is the trade, and it is made deliberately. The same discipline is what makes a fact base useful: a number without its publisher and date is a rumor with a decimal point.
Key pointThe three-part test
For any sentence about a product, ask three things: which status word applies, what date the statement was measured, and how a reader could check it without asking the vendor. If any of the three is missing, the sentence is doing marketing work, not informational work.
“First” and “only”: what can a vendor honestly say about announced connections?
A vendor can say what it announced, on what date, and what it has not done yet. It cannot say “first” without a dated public comparison that someone else could re-run.
The checkable version, with no names: a foreign hotel software provider announced connections for AI agents in September 2025, and a domestic one in October 2026. That sentence needs no ranking and cannot be knocked over by the next announcement. The category keeps moving: on 27 August 2026, Skift reported that Google began rolling out hotel booking inside its AI mode for travelers in the United States, with named partners and payment through its own wallet, and that the mode does not book on its own (Skift, 27 August 2026). A “first” written before that date would already be out of date.
“First” needs three things at once: a defined category, a dated public announcement, and a check that nobody announced earlier. Categories shift, so the claim decays while the page stays up. “Only” is harder still: it is a statement about every other company in the world, including the ones that have not announced anything.
What AI Otel says instead is a state, not a rank. The shared core is built and runs in a public demo environment, and an AI assistant can find, hold and book a room on the demo property within limits the hotel sets. The distribution hub is built with seed data, and no agency is connected yet. The question a buyer can put to any vendor: which announcement, on what date, in which category, and who else announced before you?
“Autonomous”: what does a person still confirm?
In the PMS assistant, every action sits in one of three safety tiers, and the sensitive ones wait for a person. That is why the word “autonomous” is not used: it would describe a system that decides alone at every step.
An action is one thing the assistant can do inside the hotel's own system: open a reservation, change a rate, send a message. It is a software operation, not a payment. The registry snapshot of June 2026 holds 207 actions: 107 run directly, 90 need one confirmation, 10 need two.
| Tier | What happens before the action takes effect | Actions (June 2026) |
|---|---|---|
| Runs directly | Applies without a person, and the change is logged | 107 |
| One confirmation | Waits for one person to approve | 90 |
| Two confirmations | Waits for two people to approve | 10 |
The tiers exist because the cost of a wrong action varies. A wrong internal note is cheap to correct; a message already sent to a guest cannot be recalled. The decision record follows the same logic: small changes apply automatically and are logged, larger ones wait for a person, and every change can be undone except a message already sent to a guest. This is also why the safety layer came before the model choice.
A person is not a rubber stamp in this design. The confirmation step is where the hotel's own judgment about its guests enters the system. The question for any vendor: which actions run with no person in the loop, and who is accountable when one of them is wrong?
“Certified”: which certificates exist and which do not?
None exist. AI Otel holds no security, privacy or payment certificate, so the word does not appear on the site or in this blog.
A certificate is issued by an accredited body after an audit, and it carries a number, a scope and an expiry date. A readiness review, an internal policy or a self-assessment is not a certificate, however similar the paperwork looks. What exists instead can be described plainly: each hotel's data is separated at the database level, and past records can only be appended, with changing or deleting them closed in the database.
The temptation is real, because buyers ask for certificates: they are cheap to check and easy to compare across vendors. A data-protection law is not a certificate scheme, and no company can be certified to a law in that sense. Consent, data export and retention periods are built into the data model; that is a design statement, not an audit result. The same restraint applies to proof by association: no testimonials, ratings, awards, revenue figures or funding claims appear, because none of them exist to show.
ChecklistWhen a vendor says “certified”, ask for
- The name of the standard and the certificate.
- The issuing body and its accreditation.
- The certificate number.
- The scope: which legal entity, which systems, which locations.
- The validity dates.
- Where to check it in the issuer's public register.
- Whether it covers the AI processing or only the data center.
“Tamper-evident” and “trained”: two words that stop one step short
Both words are weaker than the ones a copywriter would reach for, and both are exactly true. The stronger version of each would be a claim the product cannot support.
Tamper-evident, not tamper-proof
Tamper-evident means alteration is visible. Past records can only be appended; changing or deleting them is closed in the database, and if someone still alters one, the chain shows it. No system makes alteration impossible, and an auditor does not ask for that: the auditor asks what you would see, and when.
Trained: the honest zero
Trained means a model has been through a training process and its results measured. That has not happened here, and the product says so on screen. What does exist: the architecture, the decision infrastructure a model would run inside, and a benchmark to choose the model. The benchmark has 30 cases and 5 traps, and it has not been run. The Hotel Brain is architected and entering training; today the PMS assistant answers questions about a hotel's own property in plain language from that hotel's own data, and a person confirms anything sensitive. Model choice is settled by measured evaluation, not by a vendor name, so no model or AI vendor is named anywhere on this site. This is how AI Otel will choose a model by measurement.
NoteThe honest zero
“We have no trained model yet, and the product says so on screen. What exists is the architecture, the decision infrastructure a model would run inside, and a benchmark to choose the model.”
How do you apply the same test to any vendor's brochure?
Ask for the status, the date and the check. Three questions strip most brochures down to what is actually there, without anyone having to argue.
Status: is this live with a paying customer, built and waiting for launch, in development, designed, or planned? Ask the vendor to answer in their own words, then write the answer down next to the claim. Date and measurement: when was this measured, and what exactly was counted? A count of database tables is a measurement; “powerful” and “smooth” are not.
The check: who else can confirm it? A named customer who will take your call, a certificate number in a public register, or a demo you can open yourself without a salesperson present. Red flags worth noting: a superlative with no date, a percentage with no baseline, the word AI attached to what is really a report, and a certificate logo with no number under it.
Keep it fair. Most vendors are not lying; they write in a genre that rewards compression, and the questions are for your file, not for a fight. Run the test on AI Otel too. The fact base, the status words and the demo are there to be checked in the same way, and a claim that fails the test should be dropped, not softened. For a fuller set, see questions to ask an AI vendor before buying.
What to do next
Take the three questions with you: which status word applies, when was it measured, and how can I check it without asking you. They work on this site and on any brochure you are handed. If you want the answers for AI Otel in one conversation, request a briefing at info@ai-otel.com and bring the questions with you.
Frequently asked questions
Is AI Otel's AI real, or is it a plan?
Not as a trained model. The PMS assistant and the decision engines run today; the three brains are architected; the benchmark that will pick the model is built and not yet run. The status words, and the language rules behind them, say this plainly.
Does AI Otel hold any security or privacy certificate?
No. No security, privacy or payment certificate exists, and a readiness review is not a certificate. The controls that do exist can be described in plain words: database-level separation per hotel, append-only records, and human confirmation on sensitive actions.
Why does the blog say Built instead of live?
Because the software is working but not yet in customers' hands, and the status word is how a reader tells the difference. Live is reserved for what operates today, which is OtelCiro's client work. The five status words are defined once and used consistently across the site.
A vendor told me their AI is autonomous. What should I ask?
Ask which actions run with no person in the loop, how many there are, what happens when one is wrong, and who is accountable. Ask for the count and the date. A tiered answer with numbers is more useful than the word autonomous.