Airbnb Revenue Estimator: How Accurate Is the Number, Really?

No estimator publishes accuracy for a single address. The one vendor that publishes a listing-level band admits a 20% range. On a $60,000 estimate that is a $24,000 spread.

Not one short-term rental estimator publishes an accuracy figure for a single address. Airbtics is the only vendor that publishes a listing-level band at all, and it states that occupancy and daily rates fall "within a 20% range of actual values" in over 95% of cases. On a $60,000 revenue estimate that band is a $24,000 spread, which is the difference between a deal and a mistake.

Every other accuracy number in this industry measures something else entirely. This post shows what each one actually measures, why the tools disagree with each other by design, and the one input that changed last week and made every estimate on the internet wrong on the same line.

What an Airbnb revenue estimator actually does

It infers your bookings from a calendar it can see, because it cannot see the bookings themselves. Airbnb does not share reservation data with third parties and has no public developer API. So every tool in this category takes repeated snapshots of public listing calendars and guesses which of the unavailable nights were sold.

AirROI, which sells estimates itself, states the position more plainly than anyone: "All third-party platforms use ‘soft data’, inferring bookings from calendar changes, pricing snapshots, and review activity, since Airbnb does not share actual booking data externally."

The core problem is one sentence long, and Inside Airbnb states it outright: "The Airbnb calendar for a listing does not differentiate between a booked night vs an unavailable night." A night you blocked for your own family looks identical to a night a guest paid for. Every estimate you have ever read is a model’s opinion about which of those two things happened.

This is the gap MagicBNB was built on the other side of. Bank account integration links checking, savings, business and merchant accounts with real-time sync, and the Smart transaction ledger matches each deposit against the payout records pulled from your PMS. An estimator infers a booking from a calendar gap and can read a renovation block as demand. A cleared deposit cannot be inferred wrong. It either landed or it did not.

The accuracy claims, and what each one is actually measuring

Three vendors publish a number. None of the three measures what a buyer assumes it measures.

• AirDNA publishes "97% Classification accuracy" on its Rentalizer page. That figure describes its booked-versus-blocked classifier, which AirDNA says reads 16 different booking signals. It is not a statement about revenue accuracy, and AirDNA publishes no revenue accuracy figure anywhere.
• Airbtics publishes "over 96% accuracy" and separately states a 97% correlation between its pre-IPO prediction and Airbnb’s actual reported revenue. Both are platform-wide aggregates measured against Airbnb’s own financials.
• AirROI publishes error "typically within 1-3%" and accuracy ">95%", benchmarked against Airbnb’s publicly reported quarterly revenue.
• Mashvisor, AllTheRooms and BNBCalc publish no accuracy figure at all. BNBCalc frames its output as "market benchmarks" and states that "actual results vary by property and operations."
• Rabbu declines to give a number and says so in writing: "Some hosts outperform estimates, and some hosts underperform estimates."

Read that list again and the pattern is uncomfortable. Every published accuracy number proves the vendor can reconstruct Airbnb’s total gross booking value across millions of listings. That is a real achievement and it is irrelevant to you. Aggregate error cancels out. Your single address is where it does not.

Although we achieve platform-wide accuracy, our data may not always be 96% accurate at the listing level. In most cases (over 95%), occupancy rates and daily rates fall within a 20% range of actual values.

That is Airbtics, in its own help centre, being more honest than its marketing page. It is the only listing-level band published by anyone in the category, and it deserves credit for existing. It also tells you the honest error bar on every tool in this list, because they all work the same way.

Why two estimators give you different numbers for the same house

Because they are measuring different things and mostly do not tell you. The definitional choices below are published by the vendors themselves, and they move the answer more than the modelling does.

They disagree about what counts as revenue

AirDNA defines revenue as "the sum of nightly rates, cleaning fees, minus service fees and discounts", so cleaning fees are inside its Rentalizer projection. Rabbu excludes them entirely: "Our estimates only include income from nightly rates. The estimates do not include cleaning fees or other incidental fees." Two tools, same house, and one structurally prints a bigger number before any modelling happens.

That gap is not small. AirROI’s February 2026 analysis of 744,677 US listings found the average cleaning fee is $188 against a $288 average daily rate, and that 86.6% of US listings charge one. Whether that sits inside or outside the estimate moves a three-bedroom projection by tens of thousands of dollars a year.

They disagree about the denominator

AirDNA excludes blocked nights from the occupancy denominator entirely, counting reserved days against "active listing nights" only. Inside Airbnb refuses to use calendars for occupancy at all, instead converting reviews to bookings at an assumed 50% review rate and capping occupancy at 70%. Their published justification for choosing 50% is that it "sits almost exactly between 72% and 30.5%", which is a midpoint, not a measurement.

AirROI has quantified how much this choice matters on its own research: the definition of occupancy you pick "moves one market’s answer by 21.5 percentage points." That is larger than most of the differences operators argue about.

They disagree about seasonality

Rabbu discloses that it seasonalises using "county-level, pre-COVID demand data by month." Pre-COVID. For a 2026 projection. That is a defensible modelling choice for stable leisure markets and an actively misleading one for any market whose demand shape moved after 2020, which is most of them.

The input that changed last week and broke every estimate

Airbnb replaced the split service fee with a single host-paid fee of 15.5%, and the price-adjustment deadline for hosts outside the European Economic Area was 15 September 2026. Hosts inside the EEA and Switzerland have until 13 October 2026.

Under the old structure a host paid roughly 3% and the guest paid 14.1% to 16.5% on top. Under the new one the host pays 15.5% and the guest pays nothing. Airbnb’s own worked example is blunt: "if you keep your price at $100, you’ll earn $84.50 after the 15.5% fee is deducted and guests will see $100." Airbnb also notes that once you move, "you can’t switch back."

Now go and read almost any Airbnb profit calculator, spreadsheet template or underwriting blog published before July 2026. It models a 3% platform fee. Airbnb published the change on 7 July 2026 and updated it on 24 August. The internet has not caught up, and a 12.5 point error on the largest single cost line is not a rounding difference. It is the whole margin on a thin deal.

Worth being precise about what this does and does not break. It does not change the gross revenue an estimator projects, because estimators model what guests pay. It changes everything below that line. If you underwrote a property in the last two years using an estimate and a 3% fee assumption, your net is overstated, and no estimator will tell you.

That is the argument for underwriting somewhere you can re-run the deal rather than in a spreadsheet you rebuild from scratch. MagicBNB’s Property Analyzer takes the platform fee as an explicit input rather than a buried constant, runs purchase and lease modes side by side, and returns gross and net revenue, annual ROI, cap rate and a fixed-versus-variable cash flow breakdown with the calculation methodology written out. Every analysis stays live, so when a fee structure moves 12.5 points you reopen the deal and ask what changed instead of starting again.

What the independent research says

There is almost none, and that itself is the finding. No published study compares estimator output against a ground-truth set of actual host payouts. What exists is methodological critique, and it is pointed.

A team at the University of Glasgow’s Urban Big Data Centre, publishing in PLOS ONE, wrote that AirDNA’s methods "are not openly available for scrutiny" and that "the training data which underpin their methods is increasingly out-of-date." Their reasoning is that Airbnb stopped exposing booked-versus-unavailable in 2014, so the models were trained on a world that no longer exists.

The same paper is honest that its own daily-scrape approach hits the same ceiling: "We still cannot distinguish actual bookings from dates when the property was unavailable for other reasons." And it identifies the error direction most operators never consider. Bookings made and completed between two scrape points vanish, which undercounts. Late cancellations get missed, which overcounts. Scrape less often and both effects get worse.

What each tool costs, verified 22 September 2026

Prices below were read off each vendor’s own pricing page on 22 September 2026, because a striking amount of what circulates is dead. AirDNA’s $600 Host tier no longer exists. Mashvisor’s $17.99 Lite plan no longer exists. Both are still quoted on review sites that rank on page one.

• AirDNA: Free tier with limited Rentalizer and 12 months of history. Market Research $125 monthly or $400 annually (shown as $34/mo). Adapt dynamic pricing $20 per listing per month with a 30-day trial. Property Manager is quote-only.
• Airbtics: free income calculator with no account gate. Paid plans published as $16/month billed $195 annually, and $239/month billed $750 annually. Note those two figures are internally inconsistent on Airbtics’ own page, and its help centre separately states $119 per month.
• AirROI: core platform including the revenue calculator is $0 with no stated limit. API access is pay-as-you-go at $0.20 per estimate, halved for partners, with a $10 minimum deposit.
• Mashvisor: $49.99, $74.99 and $99.99 per month on annual billing. No free tier appears on the pricing page. Custom data exports are $35 each on Standard and $30 on Professional.
• Rabbu: free, US addresses only, monetised through agent and lender referrals rather than subscriptions.

• BNBCalc: Calculator $199 a year, Markets $399 a year, both with a 7-day trial.
• AllTheRooms: now operating under Deckard Technologies. Priced per market rather than per account, from $19 a month for a single market up to $899 for global coverage.

How to use an estimate without being fooled by it

Estimates are genuinely useful for the job they can do, which is ranking markets against each other. They are weak at the job most people use them for, which is predicting one address.

• Run the same address through three tools and treat the spread as your real confidence interval. If the three land within 10% of each other, the market is legible. If they are 40% apart, the comps are thin and no single number is trustworthy.
• Check whether each estimate includes cleaning fees before comparing them. AirDNA includes them, Rabbu excludes them. Comparing those two directly is an error, not a data point.
• Underwrite on the lowest of the three, not the average. A 20% listing-level band means the downside case is the one you have to survive.
• Model the 15.5% Airbnb fee, not 3%, on anything you are buying now.
• Never use a market-level occupancy figure as a property-level assumption. Aggregation is exactly where these tools are strong and your single door is where they are not.

Once the property is yours, the estimate stops mattering and reconciliation starts. MagicBNB’s Net Payout source of truth runs one canonical calculation behind profitability, the listings table, property detail, trends and the monthly report, so the number an owner challenges in March is the same number in September. An estimate cannot be audited. A payout can.

What replaces the estimate once you own the door

Nothing, for about ninety days. Then the estimate should never be opened again, because you have something better: your own booking history, your own cost base, and your own payouts. Most operators never make that switch. They keep checking the market estimate against a property they already own, which is a bit like reading the weather forecast for yesterday.

Here is what the handover actually looks like on a composite two-bedroom in a drive-to leisure market. The estimator projected $64,000 gross. Apply Airbtics’ own published listing-level band of 20% and the honest range was $51,200 to $76,800. Year one came in at $57,400, which is inside the band and 10.3% below the headline number the purchase was justified on.

That gap is survivable. The next one is not. Under the old split fee the Airbnb take on $57,400 was roughly $1,722. Under the 15.5% single fee it is $8,897. That is $7,175 of margin that existed in the underwrite and does not exist in the bank account, and no estimator flagged it because no estimator models host net.

This is the difference between a tool that guesses and a tool that counts. MagicBNB’s Listings table puts every door in one sortable view with net revenue, occupancy, profit, profit margin and reservation count, and health-colours the occupancy pills so the property running below 60% is visible in about three seconds. No projection, no comp set, no inference. It is the money that actually cleared, per door, arranged so the weak one cannot hide behind the strong one.

The practical sequence is worth stating plainly, because most operators run it backwards. Screen markets with an estimator, because ranking markets is the job estimates are genuinely good at. Underwrite the specific deal on your own inputs and your own fee assumptions. Then, from the day the first guest checks in, run the property on reconciled actuals and let the estimate go. If you are carrying more than two doors and still cannot answer which one earned the least last month without opening a spreadsheet, that is the gap worth closing this week, and it is worth spending an evening on a trial to see your own numbers in it rather than taking anyone’s word for it.

Frequently asked questions

How accurate is the AirDNA Rentalizer?

AirDNA publishes no accuracy figure for Rentalizer revenue estimates. The "97%" on the Rentalizer page is classification accuracy for its booked-versus-blocked model, which is a different measurement. For a listing-level error expectation, the closest published figure from any vendor is Airbtics’ statement that occupancy and daily rates land within a 20% range of actual values in over 95% of cases.

Which Airbnb revenue estimator is the most accurate?

Nobody can answer that with evidence, because no independent study benchmarks these tools against actual host revenue. Airbtics and AirROI publish the most detailed methodology, Rabbu is the most candid about limitations, and AirDNA has the largest verified-data quality layer at over 100,000 properties sharing real reservation data. Transparency is the only quality signal available, so weight it accordingly.

Is there a free Airbnb revenue estimator?

Yes, several. AirROI’s revenue calculator is free with no stated limit and shows the comparable listings behind each estimate. Rabbu is free for US addresses. Airbtics offers a free income calculator. AirDNA’s free tier includes a limited Rentalizer with 12 months of history rather than 36.

Why do Airbnb estimators give different numbers for the same property?

Mostly because of three definitional choices, not modelling quality. Whether cleaning fees count as revenue, how blocked nights are treated in the occupancy denominator, and which period the seasonality curve comes from. AirROI has shown that the occupancy definition alone moves a market’s answer by 21.5 percentage points.

Do estimators account for the new Airbnb 15.5% host fee?

Generally no, because estimators model gross guest spend rather than host net. The single 15.5% host-paid service fee replaced the roughly 3% split fee, with a 15 September 2026 deadline for hosts outside the EEA and 13 October 2026 inside it. You have to apply that yourself, and almost every calculator and template online still assumes 3%.

Can I trust an estimator for a DSCR loan application?

The lender will not use it. DSCR underwriting for short-term rentals typically runs on long-term market rent from an appraiser’s rent schedule, or on an averaged operating history, rather than on peak seasonal projections. Treat an estimator as your own screening tool, not as evidence for a lender.

What should I use instead of an estimator once I own the property?

Your own reconciled payouts, from roughly ninety days in. At that point your booking history and cost base beat any market model, because they describe your actual door rather than a comp set near it. The test of whether you have made that switch is simple: if you cannot name last month’s worst-performing property without opening a spreadsheet, you are still running on projections.

Key takeaways

• No short-term rental estimator publishes an accuracy figure for an individual address, and the only listing-level band published by anyone is Airbtics’ 20% range around actual occupancy and daily rate.
• AirDNA’s "97%" refers to its booked-versus-blocked classifier, not to revenue accuracy, and AirDNA publishes no revenue accuracy figure.
• Airbtics’ 96% and AirROI’s 1-3% error are platform-wide validations against Airbnb’s reported revenue, which says nothing about a single listing.
• AirDNA includes cleaning fees in its revenue estimate and Rabbu excludes them, so the two tools are not directly comparable on a market where the average US cleaning fee is $188 against a $288 ADR.
• Airbnb’s single 15.5% host service fee replaced the roughly 3% split fee on 15 September 2026 outside the EEA, so any calculator still modelling 3% understates platform cost by around 12.5 points of gross.
• Run three estimators, underwrite on the lowest, and treat the spread between them as the real confidence interval.
• On a composite $64,000 estimate, the honest 20% band is $51,200 to $76,800, and the fee change alone moved the Airbnb take on a $57,400 actual from about $1,722 to $8,897.
• Estimates are for ranking markets before you buy. From roughly ninety days after the first check-in, reconciled payouts beat any model, because they describe your door rather than a comp set near it.

If you want the same question asked of one vendor in depth, our breakdown of whether AirDNA is accurate goes through its classifier and its verified-data layer, and the comparison of AirDNA against real reconciled data shows what the gap looks like on a portfolio that tracks both.

An estimator tells you what a house like yours might earn, inside a 20% band, on a fee assumption that expired in September. Connect your bank and your PMS and the guessing stops: every door, reconciled, with the number that actually cleared. Start a free MagicBNB trial and see your real per-door numbers →

About MagicBNB

MagicBNB is the portfolio analytics layer that sits on top of whichever PMS and pricing tool you already run. Profitability and P&L produces a real per-property statement any day of the week, with filter modes for at-loss, low-margin, improving and highest-expense doors. Recurring rules tie utilities and other repeating costs to the right property once, then backfill past matches so the history is right too. The Deal Analyzer keeps every saved underwrite side by side and scores them against your own risk tolerance and target return, so you pick the better deal rather than the first one that looked promising. See what your portfolio earns at magicbnb.io.