Creosote is the direct-booking platform for independent hosts. It answers every guest message across Instagram, text, and email in your voice, books the ready ones on your own site, and wakes you only for the ones who need a person. 89% of ours arrive by Instagram DM, where no OTA can see them.
One property · one person · ~$2,100 collected a day since February, ~790 messages a day since March
It turns an inbound question into a direct booking on your own site. Three things it’s deliberately not:
Airbnb’s assistant answers guests Airbnb already sent you, on the platform taking the commission. It can’t read a DM, a text, or an email. All 104,742 of our messages arrived somewhere it will never see.
Those manage bookings you already have. This one works before the booking exists, on the question that decides whether it happens. A management suite is a filing cabinet. This is a salesperson.
It quotes real rates, checks real availability, escalates anything touching money, and takes the payment on your own booking site, not the OTA’s. $362,420 collected in six months, from one property, run by one person.
Pick any conversation. These are real threads from a working property, with the bot's actual replies. Note the channel tag on each one and that the guest never repeats themselves when they switch.
This is your own booking site, and the same AI from the inbox answers right on it. Guests ask here and book here, on your domain and your terms, with no marketplace in the middle.
And the moment they say yes, the checkout, still on your domain:
The checkout is one screen of a complete platform. Underneath the messaging:
This is the real thing, answering as Junegrass Ranch, the fictional property from the transcripts above. Ask about rates, dates, the animals, bringing a child. Ask for a discount. Try to cancel a booking and watch it refuse to improvise.
Everything above is the product working. Everything below is why it is a company. Open what you care about.
Any chatbot can answer "what are your rates?" These are the four moments where a booking is actually won or lost.
No unit is named that. It matched the Hayloft and answered, instead of saying it had no such property and losing the booking.
It never discounts. It reframes on what's included, then answers the real question underneath, which was whether the donkeys are friendly.
Not a question, so no reply. Most bots answer everything, and every unnecessary message is a reason to mute you.
Money, cancellations, and credits always go to a human. It acknowledges, promises nothing, and hands off. This is the rule operators care most about.
Escalations land in one queue with a draft ready to go. Approve, edit, or discard. A four-hour clock runs on each so nothing sits over a weekend.
Change a price, a policy, or a check-in time in one file and every channel is correct within a minute. No retraining, no redeploy, no prompt archaeology. It’s also what stops the bot inventing amenities that do not exist.
It does, and it’s good, inside Airbnb. It answers guests Airbnb already sent you, on the platform that takes the cut. It has never once pointed a guest to a host’s own website, and it never will.
Airbnb’s assistant exists to close bookings on Airbnb, where the commission is. It will not mention your booking page or help you own the guest. That conflict is permanent, not a gap on someone’s roadmap.
Their data model is rooms and nights. A sunset tour, a day pass, a morning feed, an add-on attached to a stay: none of it exists in their schema, so none of it can be answered, upsold, or booked.
If you do ninety percent of your volume on Airbnb and you are happy about it, their assistant covers a lot and you do not need us. This is built for the operator who wants the commission back and the relationship with it.
This isn’t a feature checklist. On calendar sync and channel management the property suites win outright, and that’s a different product. This is the guest-conversation game, the part that decides whether an inquiry becomes a direct booking, and the part nobody else is built for.
| Creosote | Airbnb’s AI | Property suites Hostaway, Guesty, Lodgify |
Chat widgets | |
|---|---|---|---|---|
| Answers on | Instagram DM, SMS, email, your site | Inside Airbnb only | Its own inbox | A website popup |
| Wins the booking for | You, direct | Airbnb | You, but only after it exists | You |
| Acts | Before the booking, on the question that decides it | At checkout, on-platform | After the booking, on operations | Before, but shallow |
| Escalates money to a human | Always | No | No | Rarely |
| Sells tours, day passes, add-ons | Yes | No | No, rooms and nights | No |
Read down the Creosote column and it is one product doing one job end to end. Read across any row and the others are each somewhere else entirely, on a platform taking a cut, or after the booking is already won, or blind to the channel where the question arrived. That is the whole opening.
Any competent team can point a model at an inbox in a weekend. What it can’t shortcut is 104,742 real messages in front of paying guests, and the year of being wrong that wrote every rule in this system.
It invented a hot tub that did not exist. It made up a discount. It opened with hedges no human says out loud. Each one is now a named rule with a regression test, checked against a corpus of real conversations before anything ships.
Money, refunds, cancellations, and group quotes always escalate. That judgement looks obvious written down and is not obvious before you have been burned by it.
Every property system is built on rooms and nights. That assumption is welded into their schema, pricing, and reporting. This one is built on tours, day passes, shared capacity, and add-ons, because it was built by someone who sells those.
Airbnb sees bookings. A property system sees calendars. Revenue tools see what got booked and at what rate. Every one of them is blind to the guest who asked the price and went quiet, because that happens in an inbox and then evaporates.
Answering the messages means holding the only record of demand that failed. Across 26,026 conversations, that’s a dataset on what objection kills a booking and which reply recovers it, and it gets sharper with every property added. It can’t be bought, and it’s worth more at a thousand properties than one.
The buyer isn’t a property type, it’s a volume: anyone drowning in guest questions who wants the booking to come direct. Experience-led places feel it first, because people have to ask before they book, but a busy standard portfolio feels it just as hard.
Strip the hospitality language off entirely and the same engine turns inbound conversation into booked revenue for any business too small to staff a front desk. Venues, tour operators, clinics, studios. Anyone whose customers ask five questions before they buy.
One ranch generates more than a full-time person’s worth of guest messaging, and 86% of it never reaches a human. At a 3% share of everything collected through it, that single property supports roughly $23,000 a year. The arithmetic from there is the whole thesis.
Pricing is a slice of every booking that goes through your checkout, counted on your own Stripe. Nothing to argue over and nothing to audit. It costs nothing in your off-season and nothing at all in a month you take no bookings.
Setup takes an afternoon: point it at your site and your listings, correct what it got wrong, connect Instagram, text, and email. It starts in draft mode, so you approve every reply until you stop wanting to.
Every figure behind this, sourced and dated, including the ones that argue against us. 71% of short-term-rental revenue now runs through three OTAs. 18% of operators get zero direct bookings. One funded competitor is already in this lane.
Read the research →