AI
Do AI assistants recommend you, and do their answers lead guests to book with you? What ChatGPT, Google's AI Overviews and others say when guests ask where to stay in your town, and what to put right first.
OneHorizyn docs. Checked against the app on .
Analytics
How often AI assistants such as ChatGPT and Google's AI Overviews name you when a guest asks where to stay in your town, and whether those answers lead the guest to book with you. The page opens on "From answer to booking", then the AI report: how often you are named, your rank, who is named in your place, where the answers' links go and, when ChatGPT is read for you, what it says you are good and not good at.
Under the report come the area's summary with what to do, links to the answers question by question, AI audit and What to fix, and, folded under "More on AI search", four tiles, the months one by one, each assistant, you against each competitor and each type of trip.
Then "Once you're on the shortlist": what each AI chat assistant tells a guest who already has you on their shortlist. Whether it picks you or each competitor, and the sentence that says so; what it says a room costs, against your prices; and your facts as it states them. The steps to booking say what it answers when asked about your facts and price directly.
- Your competitors: every share is set against them. A competitor you add is counted from the next monthly reading.
- Your guest questions (usually 50 to 80, the most valuable of them marked key) and the facts kept an eye on are chosen when your hotel is added. To change them, ask your account manager.
- For "Get your price right", your own prices must be read, as Revenue management reads them. Until they are, no price in an answer is checked.
- Your facts: the shortlist's fact questions are asked from them.
Once a month, at the start of the month, each of your guest questions is put once to each assistant read for your hotel: Google's AI Overviews, and ChatGPT and Claude where OneHorizyn has them switched on. Every answer is kept, and read for your name, your competitors' names, any other place to stay it names, and the links it gives. Where ChatGPT is read for you, it is also asked once a month what guests say you are good and not good at. A hotel just added is first read within about half an hour. The shortlist questions (you or each competitor, what a room costs, and your fact questions) are asked once a month of each AI chat assistant switched on; Google's AI Overviews are not asked them.
Until the first reading it says AI assistants have not been asked about your hotel yet, beside a card with the number of guest questions ready to ask and the competitors you will be compared with. It fills in by itself.
ChatGPT's and Claude's answers come through each assistant's own service with its web search on, which is close to, but not the same as, what its app shows a signed-in traveller. Google's AI Overviews are read from Google's own results page for each search. Each question is asked once a month, so a share from few answers, for one assistant or one type of trip, has a wide range. Gemini and Perplexity are not read for hotels yet.
- 1
Start at "From answer to booking". Each line counts the month's answers that got that far: the ones that name you, of those the ones that get your facts right, then your price, then the ones that link your own website. The biggest drop is the step where the most answers stop short of you.
- 2
Press "See the N that leave you out" under "Name you", or the line like it under a later step, to list those answers: the question, the assistant, and why each one stops there. The line then reads "Hide the N that leave you out"; press it to close the list.
- 3
To read one assistant alone, pick it on the bar at the top of the AI report; "All assistants" puts them back together, and "From answer to booking" follows the same choice. The bar is there once two or more assistants are read for you. The choice is kept in the page's address, so you can send it as a link.
- 4
Read the AI report's parts in turn: "AI visibility", "Rank", "Who is named in your place", "Sources" and, when ChatGPT is read for you, "What the assistants say about you". Each opens on one sentence with the count behind it. Press "Question by question" under "AI visibility" or "Who is named in your place" to see every guest question.
- 5
Under "Who is named in your place", "Named when you are not" lists the places named in the answers that leave you out, and one real answer shows what a guest reads, with each place picked out.
- 6
Under "Sources", see where the answers' links go: your own website, booking sites or other sites, and the sites that name your competitors and not you. How to get onto those sites is on What to fix.
- 7
Press "More on AI search" for the four tiles, how often you are named month by month and on each assistant, the table "You and your competitors" and "By type of trip".
- 8
Beside "Compared to" at the top of the page, press "Last month", "Baseline" (your first month read) or "Competitors" (the average competitor) to choose what the four tiles compare with. "From answer to booking" always compares with the month before. Until a second month is read, it says "Compared with the average competitor" instead.
- 9
Press "The answers, question by question" to read what each assistant said, "AI audit: can AI read you, and get you right?" to check your website and facts, or "What to fix, the most valuable first" to act on it.
- From answer to booking
- The month's answers followed step by step: "Asked" (every answer read), "Name you", "Get your facts right", "Get your price right" and "Link your own website". Each step counts only the answers that passed the step before, written as a count of all the answers asked. With all assistants together and two or more read, a table under the steps gives each assistant's count at every step.
- no clear change
- Beside each step's count is the range its share would most likely fall in (95 times in 100) if the same questions were asked again, and its change in points on the month before. When the change is no bigger than asking again could give, it says "no clear change" instead.
- named first in N
- Under "Name you": the answers that name you before any other hotel you are compared with. It is not a step, since an answer that names you second still sends guests to you.
- Get your facts right
- An answer passes when it states none of your facts wrongly, checked the way Your facts checks them, in what the answer says about you. An answer that states no fact of yours has nothing wrong in it.
- Get your price right
- An answer passes when it quotes no room price for you more than a fifth below your lowest or above your highest price at the latest price reading. The line under the steps gives those prices and their day, or says none are read yet.
- Link your own website
- An answer passes when its links about you include your own website. One that links only a booking site, or nothing, stops here.
- AI visibility
- How many of the answers read name you, with the range the share would most likely fall in. "By assistant" splits the count by assistant, with the place each names most, and "By type of guest" by type of trip, every type asked about.
- Rank
- Your place among every place to stay the answers name, by how many answers name each. A heading or a label in an answer ("Hotel Details", "Hotel Class") is not a place, and nor is a restaurant, a spa or a room at a hotel: "Joint" when another place is named as often, "Not ranked" when no answer names you. "On each assistant" gives your rank there, how many answers name you first, and your average place in the answer (1 is first).
- Who is named in your place
- Of the answers that leave you out, how many name another place to stay. "Named when you are not" lists the twelve places named most in those answers, and the card under it shows one real answer, shortened.
- Question by question
- Folded under "AI visibility" and "Who is named in your place". "Which assistant names you, question by question" has a square for each assistant; "Who is named, question by question" says under each question who its answers name. The figure in a square is your place in the answer; an empty square is an answer that leaves you out.
- Sources
- Of all the links in the month's answers, how many go to your own website and how many to booking and price comparison sites, which take the booking and their commission. "The sites linked most" counts the answers that link each site; "Sites that name your competitors and not you" covers the last three months.
- What the assistants say about you
- Shown when ChatGPT is read for you: once a month it is asked what guests say about your hotel, and its points are sorted into "Good at" and "Not good at", in its own words, with the day it was asked.
- Good, Losing guests
- The mark on the area's summary under the report; "Losing guests" also shows beside AI in the menu. It is "Losing guests" when fewer than 1 in 5 answers name you, when you rank below third among the places named (below the top half when four or fewer are named), or when your website turns away Google's, Claude's or ChatGPT's reader.
- What to do
- On the area's summary, when you are losing guests: the guest question that no answer names you for and a competitor is named in most often (in more than a third of its answers), with how many other questions are like it, and any reader your website turns away, each with "Open".
- AI answers that name you
- A tile under "More on AI search": the share of the month's answers that name you, with your rank among you and your competitors by how many answers name each, such as "You rank joint 1st of 9 hotels: you and your 8 competitors", and its range. "Joint" when a competitor is named in as many answers; "Not ranked" when no answer names you. Home gives the same rank.
- Links that go to your website
- Of the links in the answers that name you (links about you, or lists of the best hotels), the share that go to your own website. Higher is better.
- Answers that send guests to a booking site
- Of the answers that name you, the share whose links about you go to a booking or price comparison site and not to your website. Lower is better.
- Where you appear in the answer
- Your average place among the hotels compared, in the answers that name you: 1 means named first. "Not named" when no answer names you.
- How often you're named, by month
- The share of answers that name you and each hotel you compete with, for up to 12 months. It shows once two months have been read.
- How often each AI assistant names you
- For the month, on each assistant: you against the average competitor. Each guest question is weighed by how much it matters for bookings, so it can differ a little from the report's plain counts.
- You and your competitors
- Each hotel compared, with "Named in" (its share of the answers, with its range), "Links to its own website" and "Place in the answer".
- By type of trip
- For each type of trip, such as couples or business: how often the answers recommend you, the competitor named most and the competitors' average, then who leads and your place ("not ranked" when nobody is named). Questions are weighed as on the chart for each assistant.
How often are the answers read?
Once a month, at the start of the month. Each of your guest questions is put once to each assistant read for your hotel, and the page shows the latest month read. A hotel just added is first read within about half an hour. While a month is still being asked, "More on AI search" says how many questions are done, and the figures change when the rest are in.
Why do Google's AI Overviews name me so rarely?
Google shows an AI Overview for some searches only. When it shows none, the answer is kept as an empty one that names nobody, because that is what a guest searching sees, so it counts as an answer that leaves you out.
My share went down since last month. Is it real?
Perhaps not. An assistant asked the same question twice can answer differently, so every share comes with the range it would most likely fall in if the same questions were asked again. A change is shown in points only when it is bigger than that; otherwise the page says "no clear change".
Why does my rank differ from one part of the page to another?
"Rank" in the AI report counts every place to stay the answers name, including places you are not compared with. The tile "AI answers that name you", the table "You and your competitors" and AI's chart on Home count only you and your competitors, so those three agree. The report also counts answers plainly, while the charts for each assistant and each type of trip weigh each question by how much it matters for bookings.
How does it know an answer names a competitor?
By its name as answers write it: in full, or without the part after a comma or in brackets ("The Wrenmoor" for "The Wrenmoor, Kemp Town"), or, for a name that is one word of its own, that word alone ("Wrenmoor"), written with a capital and not as the start of a longer name, such as a street's. The hotel group or brand in brackets after a name, such as "(Seaside Group of Hotels)", still counts as that place; another area in brackets is another of the chain's places and does not. An everyday word, or a word left once your town is taken out of a name, such as "Hill" from "Brighton Hill Resort", never counts on its own, since other places are called that too.
Can I change the questions, the assistants or the competitors?
Your admins and editors change your competitors in Settings › Competitors, and every page follows them. The guest questions are chosen when your hotel is added: to change them, ask your account manager. Which assistants are asked is set by OneHorizyn, and "How this is calculated" at the foot of the page names the ones asked this month and how often.
Why is Analytics missing from my menu?
Analytics comes with AI search. When AI search is not one of your hotel's features, Analytics leaves the menu, your AI answers are no longer read and your website's AI readers are no longer checked, and a link to Analytics says so, and who can add it. The answers question by question are a feature of their own, What AI says: it comes with AI search, but can be switched on or off apart from it. AI audit and What to fix stay in the menu.
AI audit
Can AI assistants read your website, and do they get you right? The page lists checks in three groups, "Can AI assistants read your website?", "Does your website say what guests ask?" and "Does AI get your facts right?", each marked "Passes", "Partly", "Fails" or "Not checked", with a link to where it is put right.
Under the checks, once the readers have been checked, a table lists every AI reader on your website and your booking engine, let in or turned away.
- Nothing of its own. If a reader is turned away, your web team changes your website's robots.txt or server settings to let it in.
What has already been read, never checked again here: the AI readers, checked when your website is read each month; your website audit, from its latest weekly run on Tuesdays; your guest questions, judged each month; and your facts on the listings and in the assistants' answers.
"Nothing checked yet" at the top until your website or the AI answers have been read, with every check marked "Not checked". It fills in by itself.
It reads results; it does not test your website itself. Of the assistants, only Claude is asked about your facts directly, so where Claude is not read for you, "AI assistants state your facts correctly" stays "Not checked".
- 1
Read the count at the top: how many of the checks that could be made pass, and how many there are to put right.
- 2
"Can AI assistants read your website?": whether the reader behind each assistant read for you is let in, and whether your website has what they need: a robots.txt that lets them in, text that shows without scripts, and an llms.txt file.
- 3
"Does your website say what guests ask?": each type of trip, with how many of its guest questions your website answers clearly, then whether your pages open with a direct answer, have headings phrased as questions, and carry FAQ markup and structured data. Press "Open Guest questions" beside a type of trip that fails or partly passes to see which questions are missing.
- 4
"Does AI get your facts right?": each fact on the listings AI reads, and as the assistants state it. Press "Open Your facts" beside a wrong one to see where it is wrong and put it right.
- 5
Under the checks, "Every AI reader, on your website and booking engine" lists each reader as "Can read it", "Blocked by the site's rules", "Turned away" or "Couldn't check". If your website turns one away, send this table to your web team.
- 6
Everything that fails or partly passes is also on What to fix, ranked with the questions you lose.
- N of M checks pass
- The checks that pass, of those that could be made. "Partly" counts as not passing, and "Not checked" is left out of the count, so it never counts against you.
- Passes, Partly, Fails, Not checked
- Each check's mark. "Partly" is a type of trip your website answers only in part; "Not checked" is a check nothing has been read for yet.
- N of M pass
- Under each group's title: its checks that pass, of those made, or "Not checked yet".
- Your website answers N of M clearly
- For a type of trip: how many of its guest questions your website answers clearly, weakly and not at all, as Guest questions judges them. It passes only when every one is answered clearly.
- Can read it, Blocked by the site's rules, Turned away, Couldn't check
- Each reader in the table, on your website and your booking engine: let in; refused by the site's robots.txt; refused by the server; or not reached when it was checked.
- measured here
- Marks the readers behind the assistants read for you this month, such as Googlebot for Google's AI Overviews and OAI-SearchBot for ChatGPT. Only these have a check of their own above the table.
- Checked
- Under the table's title: the day the readers were last checked.
Why does a check say "Not checked"?
Nothing has been read for it yet: the readers are checked when your website is read each month, the website audit runs every Tuesday, and your facts are read from your listings and the monthly answers. A check not made is left out of the count, so it never counts against you.
Does this page check my website again?
No. It reads what was already read. When you put something right, it shows here after the next of those readings: the next Tuesday for the website audit, the next month for the readers, and the next month for guest questions, or sooner when you put a page live from Articles.
My booking engine turns the readers away. Should I worry?
Less than about your website. Booking engines often turn readers away on purpose; what matters most is that your website, where your rooms, offers and facts are, can be read.
Which readers matter?
Those marked "measured here": the readers behind the assistants read for you. Readers such as GPTBot and ClaudeBot gather pages to train AI, not to answer guests, so no check here depends on them.
What does "Partly" mean?
For a type of trip: your website answers some of its guest questions clearly, or answers some only weakly. It counts as not passing, and it is on What to fix with the rest.
What to fix
Everything that would get you into more AI answers, in one list, the most valuable first: the guest questions a competitor wins, the sites the assistants cite for your competitors and not for you, and the AI readers your website turns away. Each is ranked by the answers it touches that name a competitor and leave you out, weighed by how much each question matters for bookings.
After them come the AI audit's checks that fail or partly pass, which help every answer. Below the list are "Where competitors are recommended instead of you", with the reasons, "Sites AI assistants trust" and the page's to-dos from the Action plan; once the AI search agent has read the month's answers, its card "When AI picks a competitor instead of you" sits under the list too.
This month's AI answers (the last three months of them for the sites to get onto), the AI audit, your busy periods for what is coming, the AI search agent's reading and the Action plan's to-dos. Nothing more is read for it.
Until the first reading of the AI answers it says AI assistants have not been asked about your hotel yet. After that, "Nothing to fix this month" means no question, site or reader touches an answer you lose, and no check fails.
It ranks by the answers you lose; it does not guess at answers won or bookings. The reasons a competitor is chosen point at causes; they do not prove them.
- 1
Start with "The most valuable first": the ten that touch the most answers you lose to a competitor, numbered. Each says what kind of fix it is, how many guest questions it touches, and why. "Show the other N" under them opens the rest.
- 2
For a question you lose that your website does not answer, press "Answer it on your website": Articles opens on the page about its type of trip, which answers every open question of that type. For a question about choosing between hotels, such as "What is the best hotel for a wedding in Jaipur?", which no hotel's own page can answer, press "Pitch the article AI cites" when the assistants cite an article that lists or compares places to stay for it, or "See how to win it back" when they cite none.
- 3
For a site to get onto, press "Open the page AI cites" to see the page the assistants link for your competitors, then get onto the site the way "Sites AI assistants trust" says.
- 4
Under "These help every answer", put right what the AI audit found, such as a type of trip your website answers only in part or a wrong fact on a listing. Each leads to the page where it is done.
- 5
Under "Where competitors are recommended instead of you", press a question on the left to see on the right why the competitor is chosen, what it has that you do not, and what to do.
- 6
Under them, "How to win it back" has the fix for that question. For a question your own website can answer, "Write the page" opens Articles on the page about its type of trip; once you press "I've published it" there, this card says, assistant by assistant, how many questions of that type each one names you on since it went live, and the results after 30, 60 and 90 days, measured as your Action plan measures a done to-do. For a question about choosing between hotels, it names the article the assistants cite most for it that lists or compares places to stay, such as a "best hotels in …" list or a guide to where to stay, with an email to its writer: press "Copy the email", and send it to the writer named on the article or through the site's contact page. A news story, a list of bars or restaurants, a page about one hotel, a site's front page, a wiki such as Wikipedia or Wikivoyage, and a forum are never offered, as no writer there chooses hotels; when no such article is cited, the card says so.
- 7
Press "Check again now" to ask the question again of each assistant read for your hotel, at once, and see where you stand today: which assistant names you and in what place, the competitors it names, and the sites it cites. It costs a few cents, so each question can be checked once a day and up to 10 a day in all. What it finds is kept apart from the monthly readings, so the month's figures stay as they were read. It is for admins and editors, and not for the demo hotel. Where it can't be pressed, the card says why: on the demo hotel, for someone who can only look, or when no assistant that can be asked again is set up for your hotel.
- 8
Under "Sites AI assistants trust", see the magazines, guides, lists and listings the assistants cite for your competitors and rarely or never for you, with how to get onto each.
- 9
Once you have put something right, the next monthly reading shows whether the assistants name you more, and What changed says so for a page about a type of trip in the three months after it went live.
- Start with these 10
- The headline when more than ten things are tied to guest questions; otherwise it says how many there are to fix, or "Nothing to fix this month". Under it, how many there are in all, and how many are tied to guest questions.
- Question you lose, Site to get onto, Reader turned away, Check that fails
- What kind of fix each row is: a guest question a competitor is named for more often than you; a site the assistants cite for your competitors and not for you; an AI reader your website turns away; or a check that fails or partly passes on AI audit.
- touches N guest questions
- The guest questions whose lost answers the fix touches. A lost answer names a competitor and leaves you out.
- N questions a competitor wins
- The questions where a competitor is named in more answers than you and the question matters for bookings, the most important first. Each shows its type of trip, the share of answers that name you, the competitor's share, and "busy season coming" when it is.
- Why … is chosen
- For the question picked: what the data shows the competitor has and you do not, such as more reviews, a higher rating, more sites that mention it, or a page about that type of trip, each with "You", the competitor, and "What to do".
- names you on N of M questions about …
- In "How to win it back", once a page about that type of trip is live: for each assistant, how many of the type's questions it names you on in the first monthly reading after the day the page went live, and how many before it ("up from", "down from" or "as before it went live"). A month's reading is taken at its start.
- After 30 days, After 60 days, After 90 days
- The share of the answers about that type of trip that name you in the month 30, 60 and 90 days after the page went live, against the month before it, worked out as your Action plan works out a done to-do's results. One still to come says which month's reading measures it.
- Checked again on …
- What "Check again now" found that day: for each assistant, the place it names you in, or that it leaves you out, the competitors it names first and the sites it cites, or that it could not be asked.
- Cited for competitors, Cited for you
- In "Sites AI assistants trust": how many times over the last three months the answers cite each site for your competitors, and for you ("never" when not at all). A site is listed when it is cited at least twice for competitors and at most a fifth as often for you.
- How to get on it, It helps now
- How a hotel gets onto that kind of site (pitch the editor, complete a listing, earn it through ratings and reviews, or take part), and the competitors it helps today.
- When AI picks a competitor instead of you
- Once the AI search agent has read the month's answers: for each competitor it looked at, how often the answers name it and you, the reasons it found, with the answers' own words quoted where it has them, the sites cited, and what to change.
- What to do
- At the foot: up to four of the Action plan's to-dos for the AI pages, Guest questions and Your facts, most impact first, with their impact ("High", "Medium" or "Low") and where each stands.
How is the list ordered?
By the answers you lose that each one touches. An answer is lost when it names a competitor and leaves you out. Each counts as its share of its question's answers, times how much the question matters for bookings, more when its busy season is coming. What cannot be tied to questions follows, in the audit's order.
What does "busy season coming" mean?
A busy period on Busy periods for that type of trip is on now or starts within the next 120 days, so the question counts for more.
Why is a question I already answer on my website listed?
Because a competitor is still named more often for it. Its link then says "See why" the competitor "is chosen" rather than "Answer it on your website", and the reasons are under "Where competitors are recommended instead of you".
Will fixing these bring bookings?
The page does not promise it. It ranks by the answers you lose today and does not guess at answers won or bookings. The reasons a competitor is chosen are what the data shows about both hotels; they point at causes and do not prove them.
When does the list change?
With the monthly reading of the AI answers. The part from the AI audit changes as your website and your facts are read again.
What AI assistants say
Each key guest question as a card, with a column for each assistant read for you: whether it recommends you, how early in its answer it names you, the site it relies on, and a short part of its answer in its own words. It sits under Analytics in the menu: open it with "The answers, question by question" on Analytics, or with the number of guest questions at the foot of every page, which shows them all.
The same monthly answers as Analytics. Every answer is kept, so each one here can be checked.
"Nothing collected for this yet" until the first monthly reading, which fills it in by itself.
One short part of one answer is shown for each assistant; every answer is kept. The site after "Relies on" is the first link about you, so it is not shown when the assistant leaves you out.
- 1
Read each key question's card: for each assistant, whether it recommends you and in how many of its answers, how early it names you, the site it relies on, and its own words.
- 2
Press "Show all N questions" to see every guest question, not only the key ones; "Show the key questions only" goes back.
- 3
Press the site after "Relies on" to open the page the assistant used.
- 4
Open "How this is calculated" at the foot of the page for how "Recommends you" is decided.
- Recommends you, Sometimes recommends you, Doesn't recommend you
- Whether the assistant names you in its answers to the question: in 67% or more of them, in some of them, or in none. Beside it, how many of its answers name you.
- Usually names you first
- How early it names you, on average, among the hotels compared: first, or around 2nd, 3rd and so on. When it does not name you, "Names … first" gives the hotel you compete with that it names first most often, or "Names no hotel we track".
- Relies on
- The site the assistant links about you, or else the first site in an answer that names you, with what kind of site it is. It shows only when the assistant names you.
- The words under it
- A short part of one of its answers, around where it names you, or its opening when it leaves you out, with links and formatting taken out.
- Not asked yet
- That assistant was not asked this question in the month shown.
Which questions are the key ones?
The most valuable of your guest questions, marked when your hotel was added: usually about 12. "Show all N questions" shows every one, and so does the number of guest questions at the foot of every page.
Why is it nearly always "Recommends you" or "Doesn't recommend you"?
Each assistant is asked each question once a month, so it either names you or does not. "Sometimes recommends you" needs more than one answer to the same question.
Why does a column say "Not asked yet"?
That assistant was not asked this question in the month shown. Analytics' "How this is calculated" names which assistants were asked, and how many questions each.
Who can see these answers?
Everyone who can open your hotel in OneHorizyn. Nothing here is shown outside it.
Agents
"Your agents": the nine agents that gather what is public about your hotel and the hotels you compete with, one card each. Seven use AI to read, explain and write; the others are automatic checks. Each card says what the agent does, when it last ran and what it found, and leads to the page it fills. It covers every area, so it is in no part of the menu: open it from these docs, or at its address, /agents.
Each card is read from what its agent last collected for your hotel. Its foot says how often it runs and what it reads.
"Not run yet" and "Nothing collected yet." on a card until its agent first reads something for your hotel.
The AI parts work only while the AI is switched on for OneHorizyn, except the Review agent's, which can also run on OpenAI. Until then the other agents are automatic checks alone, and the page is open only to OneHorizyn's own team.
- 1
Read each agent's card: what it does, what its AI part does (after "AI:"), when it last ran and what it found, and at its foot how often it runs and what it reads.
- 2
Press a card to open the page it fills: the AI search agent's opens Analytics.
- 3
Under "Built on what the agents find", "Your monthly briefing" opens Home, where it is written, and "Ask OneHorizyn" says where to ask a question about your hotel.
- Last run
- When the agent last collected for your hotel: a day, or a month for the monthly readings. "Not run yet" before its first.
- AI read it
- When its AI part last read what was collected, for the agents that have one.
- What it found
- One sentence from your data, such as how many reviews are waiting for a reply or how often the AI answers name you. "Nothing collected yet." before its first run. When the nights the Proof agent last photographed have all passed, it says how many screenshots it took that day, for which nights, and that those nights have passed.
- AI:
- What the AI does in that agent. An agent without this line is an automatic check.
- Every month
- At the foot of a card: how often the agent runs, then what it reads. For prices and reviews it is the full reading; both are also read every week.
- Your monthly briefing
- "Written" with its day once the AI has written it on Home, or "Written after the next monthly refresh".
Why can't I find Agents in the menu?
Since 12 October 2026 it is in no part of the menu, because it covers every area. Open it with the button at the top of this part of the docs, or at its address, /agents.
Why does a card say "Not run yet"?
Its agent has not collected anything for your hotel yet. An agent whose feature your hotel does not have collects nothing, so its card says "Not run yet" until the feature is added, or keeps its last run from before.
Who can open it?
Everyone who can open your hotel, while the AI agents are on for it. While the AI is switched off for OneHorizyn, only OneHorizyn's own team can open the page, and it says so at the top; everyone else, your group's admins included, is sent to Home.
Does "Every month" mean my prices and reviews are a month old?
No. It is the full reading. Prices and new reviews are also read every week, and screenshots are taken every Monday: see When things are read.
What changed
Analytics' tile ranks you among you and your competitors as Home does, and says "joint" when a competitor is named in as many answers.
An answer that names a competitor with its hotel group in brackets, or by the one word of its own name, counts it; a common word left once your town is taken out of a name never does.
What to fix says "Pitch the article AI cites" only for a question whose answers cite an article that lists or compares places to stay, and "See how to win it back" for the rest.
"How to win it back" offers an email only about an article that lists or compares places to stay, never a news story, a list of bars or a Wikivoyage page, and says so when the assistants cite none.
Analytics no longer counts a heading or a label in an AI answer, or a restaurant, spa or room at a hotel, as a place to stay, and "By type of guest" shows every type of trip.
"How to win it back" never offers an email about Wikipedia or a page about one hotel, says why "Check again now" isn't there when it can't be pressed, and What to fix counts the questions answered only weakly among those with no clear answer, as AI audit and Articles do.
The Website agent counts the guest questions your website can answer, as Guest questions does, and the Proof agent says what it took when the nights it checked have passed.
What to fix has "How to win it back" for each question you lose: the page about its type of trip to write, or an email to the writer of the article AI cites; once your page is live, what it changed for each assistant and its results after 30, 60 and 90 days; and "Check again now", to ask a question again of each assistant at once.
What AI tells a guest who has you on their shortlist: you or each competitor, your price and your facts.
Analytics opens on how far each AI answer takes a guest towards booking you, step by step.
Each share of AI answers is shown with its range, and a change on last month is called only when it is real.
AI is three pages: Analytics, AI audit and What to fix.
Agents is no longer in the AI part of the menu, since it covers every area: open it from the docs or at /agents.
Each assistant read for your hotel is asked every one of your guest questions each month, ChatGPT included, not only the key ones.
When AI search is switched off for a hotel, Analytics leaves its menu, a link to it says so, and its AI answers are no longer read.
AI search says which assistants it read, with ChatGPT beside Google's AI answers.
