hostAI Research · June 2026

The Direct-Fit Profile

We analyzed twelve months of booking data across 5,500 short-term rental listings to find out what actually predicts direct booking performance. Most of the usual explanations, including price, size, and amenities, did not survive the test. The signal that held up is one the industry rarely quantifies: a listing's review count, which appears to track how heavily its demand runs through the OTAs.

A hostAI research report · Based on 350,000+ bookings and $298M in booking value

Direct booking has been a strategic priority for short-term rental operators for two decades. Over that time, the industry has accumulated a set of beliefs about which properties are suited to direct and which are not. Premium properties win direct. Certain markets are direct-friendly and others are not. Bigger brands have the advantage. Review-rich listings convert better.

Most of these beliefs have never been tested at scale. We decided to test them.

This report summarizes what we found when we analyzed twelve months of booking data across more than 5,500 listings: which property attributes are associated with higher direct booking contribution, which widely held assumptions did not hold up in our data, and what we still do not know.

One note before the findings. This is research drawn from our data pipeline, not a census of the industry. The sample skews toward operators who have shown interest in direct booking infrastructure, and toward a specific mix of markets. We have noted the limitations throughout, and we intend to repeat and refine this analysis annually as the dataset grows.

01The baseline picture

We analyzed paid bookings created between May 2025 and May 2026 across 5,509 listings managed by 138 property management companies. We included listings with at least 20 bookings in the window and measured direct contribution at the listing level. Because listings run by the same company tend to behave alike, our comparisons account for that rather than treating every listing as independent; the technical details are in the methods note at the end.

Across the full dataset, OTAs account for roughly 87 percent of booking value. Direct channels account for the remaining 13 percent, split between online direct (an operator's own website and booking engine) and non-online direct, a residual bucket that includes phone, repeat guests, and other relationship-driven bookings but is noisier than online direct and likely also captures manual entries and PMS quirks.

Where the money comes from

The channel mix across 5,500 listings

Share of total booking value, all listings, May 2025 – May 2026

02What we tested

We examined the attributes operators most often point to when explaining direct booking performance: price tier, property type, bedroom count, guest capacity, market and location, operator portfolio size, time in market, average rating, review depth, and a set of specific amenities including pools, hot tubs, pet-friendliness, EV chargers, and dedicated workspaces.

For each, we asked the same question: does this attribute still predict direct contribution once everything else is held constant? Many attributes that look meaningful on their own turn out to be standing in for something else. The test that matters is whether an attribute survives when the others are accounted for.

03What did not predict direct contribution

This is, in some ways, the most useful part of the analysis.

Bigger operators did not do better. If anything, they did worse. One of the most common beliefs in the industry is that direct booking is a game for large, well-resourced brands. The data does not support it. The largest portfolios in the dataset, those managing more than 100 listings, captured the lowest direct share of any size band: roughly 4.6 percent online direct and 10.2 percent total, both below the dataset baseline. Smaller operators consistently did better. Scale, on its own, does not appear to buy direct booking performance.

Price tier did not survive a fair comparison. The highest-priced listings show somewhat higher direct contribution at first glance, but the difference disappears once you compare properties of similar type in similar markets. Expensive properties do not appear to win direct because they are expensive.

Bedroom count and guest capacity did not survive it either. Once like is compared with like, neither property-size measure predicted direct contribution on its own.

Most amenities did not survive it. Pet-friendliness, hot tubs, EV chargers, and workspaces all looked different at first glance, but none of those differences held up once similar properties were compared with each other. The lone exception was pools, which kept a positive association of roughly three to five percentage points.

Property type separates listings, but only descriptively. Format is the one attribute in this group that does carry some signal, though we treat it as texture rather than a headline because the effects vary by metric and we did not put property type through the same battery of tests as the review finding below. The pattern is also two patterns at once. Private rooms and cabins over-index on online direct, the channel that runs through an operator's own site. Villas do something different and more striking: 32 percent of villa booking value comes direct, more than double the dataset average, but almost all of it through non-online channels, repeat guests and relationship-driven rebooking rather than online discovery. Meanwhile the three largest categories by volume, condos, homes, and townhouses, sit at or below baseline, with homes alone making up the single biggest slice of booking value and pulling the average down. One note on definitions: property types come from each listing's classification in its source PMS or OTA export, and a residual set of listings without a clean type match is held out of these figures.

04The signal that survived

One attribute kept predicting direct contribution after the others fell away, and it runs against industry intuition. In the raw data, listings with just 1 to 9 reviews captured far more of their revenue through direct channels than listings with 100 or more. The relationship is steady across every review band in between: the more review history a listing accumulates, the smaller its direct share tends to be.

A note on terminology: the review count here comes from listing data where Airbnb identities are present, so it is best read as review count primarily on Airbnb, not a verified count across every OTA.

The raw pattern

More reviews, less direct booking

Direct share of booking value, by review count per listing (primarily Airbnb). Raw shares, before any like-with-like adjustment.

Online direct Total direct (incl. non-online)

This raw pattern is real, but it would be a mistake to read it at face value. A low review count travels with several other things: low-review listings tend to have fewer total bookings, shorter time in market, and longer average stays. Each of those could drive direct share on its own. A skeptic should ask whether this gap is really about reviews at all. So we tried to explain it away, three times, each test stricter than the last. The chart below shows what survived.

The first test makes the comparison fair: instead of comparing every listing against every other, it compares listings of similar type, similar price, and similar market. The second test rules out three specific suspects. Maybe low-review listings are just newer, or take fewer bookings, or host longer stays, all of which would mean fewer reviews regardless of anything else, so we compared listings that are also similar on those three things. One suspect deserves special mention because it is mechanical: direct bookings do not generate OTA reviews, so a listing doing lots of direct business collects fewer reviews almost by definition. We checked this directly by measuring reviews per booking and reviews per month rather than raw counts, and the pattern held. The third test is the toughest. Maybe this is about companies, not properties: some operators are simply good at direct across everything they manage. So we stopped comparing across companies entirely and looked only within each operator's own portfolio, their low-review listings against their own high-review listings. Same company, same website, same marketing, same team.

How much survives

The gap shrinks under scrutiny, but does not vanish

How many more points of revenue listings with 1 to 9 reviews book direct, versus listings with 100 or more, as the comparison gets stricter

Online direct Total direct (incl. non-online)

What is left after all of that is modest but real, and it is steadiest for online direct. Comparing a single operator's own listings against each other, the ones with light review histories still capture meaningfully more online direct. We want to be careful about what that means. This is a correlation, and the analysis cannot establish what causes what. The most we will say is the interpretation the data is consistent with: deep listing-review accumulation appears to mark OTA-heavy demand patterns more than it marks portable trust. Some of that reflects exposure and stay length. Some of it may reflect operators who built their demand on the OTAs and never developed the direct muscle. We cannot yet separate those, and we say so plainly in the next section.

05Why the review pattern? Four explanations

The review-depth finding is the one most likely to be misread, so it deserves a careful accounting. A low review count travels with several other things, and each could generate the pattern on its own. We tested four explanations. None is fully ruled out, but the data weighs them differently.

One: a mechanical artifact. Direct bookings produce no OTA review, so a high-direct listing accumulates fewer reviews almost by definition. This is real and it explains part of the gap. But measured as reviews per booking and reviews per month, low review counts still predict more direct, so the artifact does not account for the whole finding.

Two: length of stay. Long-stay properties turn over fewer times, so they collect fewer reviews, and their guests, who research more and save more by booking off-platform, may lean direct. The data supports this as a partial driver: long-stay listings do over-index on direct, low-review listings are disproportionately long-stay, and controlling for stay length softens the gap. It does not erase it.

Three: the OTA ranking flywheel. The intuitive story is that well-reviewed listings rank highly, fill their calendars through the OTAs first, and leave only leftover inventory for direct. The testable version of this did not hold up. On high-review listings, direct bookings are not systematically later-booked or lower-priced than OTA bookings, which is what residual-inventory capture would look like. We cannot test true calendar occupancy without availability data, so this is not fully closed, but the evidence we have does not support it.

Four: genuine OTA dependency. Operators who build their demand on the OTAs may never develop the direct muscle, brand, or repeat-guest base, leaving review depth as a marker of structurally outsourced demand. This is the explanation that best fits what remains: the gap survives every fair-comparison test above, including the within-portfolio one, for online direct. But the mechanism itself, repeat direct guests escaping OTA dependency, was not directly testable in this dataset, so it remains plausible rather than proven.

The honest read across all four: light listing-review accumulation marks OTA-heavy demand patterns more than it marks portable trust, and that association is not explained away by exposure, stay length, or operator mix. What we cannot yet say is that reviews themselves cause the dependency. That distinction is the difference between a research finding and a sales pitch, and we are going to hold the line on it.

06What this might suggest

Two observations follow, each offered with appropriate caution.

First, the reasons operators most often give for weak direct performance, namely price point, property size, and portfolio scale, did not show independent predictive power here. Operators who have concluded that direct is structurally unavailable to them may be working from weaker evidence than they assume.

Second, if even part of the review-depth pattern reflects demand dependency rather than mechanics, it points to a cost of OTA reliance that appears on no P&L line. Every booking taken through an OTA deepens an asset that lives on the platform. Our data is consistent with the possibility that this accumulation works against the direct channel over time, though, to be clear one more time, this is a correlation, and establishing cause will take longitudinal work we have not yet done.

07Limitations

Several limitations should frame how these findings are read. The sample reflects operators in our data pipeline and over-represents those with an existing interest in direct booking; absolute contribution figures should not be generalized to the whole industry, though relative patterns within the dataset are more robust. The analysis is cross-sectional: it observes listings at a point in time and cannot track how direct contribution evolves as attributes change, which is why we are careful not to claim that reviews cause dependency. The review-depth figures above are reported after the toughest test we ran, the within-portfolio comparison, rather than before it; that comparison is solid for online direct and right at the edge of confidence for total direct, and total direct also includes the noisier non-online residual bucket. One related result worth stating plainly: time in market did not explain the review finding away. Longer-tenured listings, if anything, do somewhat more direct, while high review depth does less, which is the opposite of what a simple age story would predict. Geographic coverage skews toward a specific set of markets across North America, Europe, Latin America, and Southeast Asia.

08What comes next

We plan to repeat this analysis annually, with a larger dataset, a fuller market taxonomy, a repeat-guest layer that would let us test the dependency mechanism directly, and longitudinal tracking that can begin to separate correlation from cause. If the review-depth pattern holds across future editions, it would have real implications for how operators think about channel strategy, and we would rather establish that carefully than claim it prematurely.

The honest summary of this first edition: the attributes operators are told matter mostly did not, and the attribute the industry rarely quantifies, listing-review depth, carried a real signal that survived every attempt we made to explain it away.