AI Discovery Black Book
Benchmarks, trends, and findings from the fastest-growing source of organic leads
We analyzed 3,811 top competitors across five and a half months to figure out what makes AI discovery engines recommend one apartment over another.
Peek | June 2026 | Last updated June 2, 2026
of all AI citations come from ILSs like Zillow or Apartments.com
top website vendor for AI visibility
of those ILS citations come from the Zillow network (Trulia, Redfin, Realtor)
of top 10 NMHC managers outperform the average community
What's changed since March
The first industry report on what actually drives AI visibility in multifamily
We dropped the first AI Discovery Black Book in March 2026. We'd planned to refresh it quarterly, but the market is moving fast enough that we pulled the second edition forward. It's built on nearly double the data, across five and a half months, with a sharper methodology behind the rankings.
Key Findings
Seven takeaways on how AI engines recommend multifamily properties, plus what's shifted since the first report dropped in March. Throughout, we use AI Discovery Score (the share of relevant AI queries where a property shows up) as our core visibility metric.
Two data sources still control AI visibility, but Zillow's lead narrowed to a near-tie.
In March, Zillow and its syndication network powered 57% of all ILS citations, a 28-point lead over CoStar. Today, that lead has shrunk to just 7 points. A Zillow-only strategy now misses more than half the data pipeline that AI engines use to recommend apartments.
You need a different strategy for each LLM, and Claude is the biggest visibility upside.
Perplexity is the most ILS-dependent at 81% of its citations. Gemini sits at 73%, ChatGPT 65%, Claude just 55%. So you can't optimize for all four LLMs the same way. Claude is also citing 47% more unique communities per response than six months ago, making it the easiest LLM to break into.
Top performers are investing in their property websites more than everyone else.
Top performers' citations come disproportionately from their own community websites, not listing services: our best single community drew 249 mentions from its site alone. Claude leans hardest on property site content, and these teams earn it - real content, not template pages.
Properties underperform significantly when missing pricing and commute data.
Top performers outscore assessed properties by 30% on price queries and 26% on commute. About 70% of properties still miss both. Strong price and commute content drives way more AI discovery lift than amenity descriptions do.
Your website vendor matters, but it's not the only lever.
Measured by MAAR (Market-Adjusted Appearance Rate), Repli leads, followed by Apartments247, Jonah Digital Agency, Entrata, and RentCafe. Vendor choice matters, but using a template is not enough. Pair a strong vendor with deep, structured content (real photos, real prices, real amenity tags) and you get both levers working in your favor.
AI discovery is earned at the community level, not the portfolio.
Scale is neither necessary nor sufficient. The NMHC top-10 managers average a 1.3% AI Discovery Score against a 14.6% benchmark, and not one beats the average community. Listing accuracy, content depth, and ILS coverage at each property determine who LLMs recommend, not sheer scale.
Broad ILS coverage matters more than picking the right platform.
Zillow's network now extends through Redfin (via the $100M February 2025 deal) and Realtor.com (via March 2024 syndication). A listing on Zillow automatically reaches all three. Distribution breadth beats single-platform optimization. Pair Zillow's network with a CoStar presence (Apartments.com, Homes.com) and you cover ~85% of the ILS pipeline AI sees.
Want to see how your properties stack up?
Pull a free report on any property to see how you rank against local competitors.
The Data Landscape
68% of all AI citations come from ILS platforms, down from 71% in March. Two parent companies control 84% of that pool. Breadth across data sources matters more than breadth across platforms.
When someone asks ChatGPT or Gemini to recommend an apartment, where does that answer come from? Mostly ILS platforms. But the platform names are misleading. Many of these sites share the same underlying database thanks to syndication deals.
The Zillow pipeline isn't the default anymore. Of the citations AI pulls from ILS platforms, Zillow-sourced sites supplied 57% in March but just 45% today, a 12-point slide. CoStar climbed from 28% to 38% of that ILS pool over the same window. The slide paused in May. June will tell us if Zillow stabilized or CoStar keeps closing the gap.
ILS citation share over five and a half months
Data as of May 14, 2026
Syndication Network
| Data Source | Original Report (Dec 2025–Feb 2026) | May 2026 | Δ | Composition | Status |
|---|---|---|---|---|---|
| Zillow-Sourced | 57.2% | 45.2% | −12.0 pts | Zillow Group (Zillow/Trulia/HotPads/StreetEasy) + Redfin/Rent.com/ApartmentGuide (syndicated) + Realtor.com (syndicated) | Owned + Exclusive deals |
| CoStar Network | 27.5% | 38.4% | +10.9 pts | Apartments.com, ApartmentFinder, ForRent, ApartmentHomeLiving, Homes.com | Owned |
| Yardi Systems | 10.4% | 9.2% | −1.2 pts | RentCafe | Owned |
| Independent | 4.9% | 7.2% | +2.3 pts | Apartment List, Zumper | Independent data |
What this means: 45% of ILS citations feeding LLMs come from one listing database, Zillow's. If Zillow ever changes how it structures data, licenses it, or lets bots crawl it, nearly half the apartment data AI sees could vanish overnight.
FTC lawsuit update: the FTC and five state AGs sued Zillow and Redfin in September 2025, claiming the $100M deal kills competition. As of May 2026, the deal's still active and Redfin still mirrors Zillow exclusively. If a court ever unwinds it, the Redfin-owned network's 18.1% of ILS citations (Redfin, Rent.com, ApartmentGuide) would redistribute back to independent sources. We're watching.
Why breadth across data sources wins
How many distinct data sources you cover matters more than how many ILS platforms you're on. Zillow, Trulia, and Redfin look like three-platform coverage, but all three pull from the same database, so you're only adding surface area to a single source. To diversify, add the CoStar network (Apartments.com, ApartmentFinder, Homes.com) and Yardi's RentCafe, which run on independent databases.
Takeaway: Cover different data sources, not more platforms. Three Zillow-syndicated listings don't diversify you, they just stock you on more shelves of the same store.
How each AI model plays it differently
Not all AI models treat your listings the same way. Claude gets 55% of its citations from ILSs and Perplexity gets 81%. So the playbook for each has to be different.
And they're pulling apart faster every month. The four models differ on three things that matter: how much they lean on listing services, how much they concentrate on a few top properties, and what content they reward. A property optimized for one can be nearly invisible on another.
ILS share by AI model
Share of competitor citations that come from ILS sources, by model. Citation-weighted average across the four is 68%.
Data as of May 14, 2026
How evenly each model spreads recommendations
Some models concentrate their recommendations on a few top properties; others spread them more evenly. The gap below is how far ahead a model's top properties sit from a typical one. A big gap means the model keeps surfacing the same few winners; a small gap means scores are bunched closer together. ChatGPT and Claude stay the most even. Gemini and Perplexity reward their top properties the most, though Gemini is evening out fast.
| Provider | Mar 2026 Gap | May 2026 Gap | Change | Status |
|---|---|---|---|---|
| Gemini | +41% | +22% | −19 pts | Top-heavy, easing fast |
| Perplexity | +36% | +29% | −7 pts | Most top-heavy |
| Claude | +13% | +15% | +2 pts | Fairly even |
| ChatGPT | +11% | +7% | −4 pts | Most even |
Claude is the easiest model to break into
Claude recommends 47% more unique communities per response than it did six months ago, the widest variety of any model. ChatGPT spreads its scores evenly too, but Claude recommends far more different properties in the first place, so it's the best place to break in if you're not already a name. It also weights property website content more heavily than ChatGPT, Gemini, or Perplexity, so a strong, well-structured site is disproportionately likely to surface there.
If you're betting on ILS alone, Claude is already leaving you behind, and ChatGPT's source mix is shifting under your feet. Claude pulls just 55% of its citations from ILSs, the lowest of any model. ChatGPT halved its Zillow.com citations and nearly doubled Apartments.com, with the difference flowing to property websites and PMC corporate pages.
Most operators don't know which models their renters use. Discover tells you, then gives you the right play for each:
- ChatGPT or Claude: invest in property website content and structured, machine-readable data.
- Perplexity: prioritize CoStar network coverage (Apartments.com, ForRent, Homes.com).
- Gemini: invest in broad ILS distribution. Content investment matters less.
Want to see how you're performing on Gemini?
Pull a free Gemini report on any property. If you want to learn more about the other platforms, reach out to our team.
What content actually moves the needle
Price and commute content still drive the biggest AI Discovery Score lift. About 70% of properties still miss both. But top performers' lead is shrinking every month.
Priorities haven't changed since March. What's shifted is the urgency.
The window to differentiate on price content is closing. In March, strong price content gave a property a +46% AI Discovery Score lift over everyone else. Today that's +30%. Commute fell from +35% to +26% over the same window. Same fix, smaller reward.
AI Discovery Score lift: then → now
Lift = the gap between top performers and everyone else on each content category.
| Category | Mar 2026 Lift | May 2026 Lift | Change | Status |
|---|---|---|---|---|
| Price | +46% | +30% | −16 pts | Closing fast |
| Commute | +35% | +26% | −9 pts | Closing |
| Unit | +29% | +22% | −7 pts | Closing |
| Location | +22% | +13% | −9 pts | Closing |
| Amenity | −4% | −6% | −2 pts | Not a differentiator |
Lift trajectory by category
Data as of May 14, 2026
Takeaway: Strong price and commute content is the biggest lever you can pull right now, more than amenities or lifestyle copy.
Which owners and operators are winning in AI discovery?
The PMCs winning at AI discovery share common traits. Portfolio size isn't one of them.
We looked at which management companies show up in AI discovery and which don't. Twice the data made the answer sharper. Zero of the NMHC top-10 managers outperform the average community, in every month of the dataset. The fastest climbers run the full size range, from a few thousand units to six-figure portfolios, and that's the tell: AI discovery rewards execution at each individual property. Owning more doors buys you nothing.
Top PMCs: current 5-month leaderboard
| Owner | Top-5 Appearances | Median MAAR | Portfolio (units) |
|---|---|---|---|
| Morgan Properties | 63 | 73.7% | 96k |
| Veris Residential | 57 | 71.0% | 12k |
| UDR | 52 | 64.3% | 60k |
| Equity Residential | 53 | 61.8% | 83k |
| AvalonBay Communities | 56 | 60.5% | 98k |
| Connor Group | 19 | 48.3% | 15k |
| GID | 15 | 46.4% | 49k |
| Nuveen Real Estate | 29 | 46.0% | 51k |
| Continental Properties | 18 | 44.5% | 26k |
| Mill Creek Residential | 26 | 42.6% | 49k |
| Berkshire Residential | 31 | 37.6% | 36k |
| LivCor | 37 | 36.4% | 193k |
This table is ranked by Median MAAR (defined below), not appearance count. Portfolio shows the operator's total Yardi-tracked units: the winners span 12k to 193k.
Why we built MAAR
The more data we gathered, the clearer it became that simply counting how often a property shows up isn't a fair way to compare properties across different markets. Start with the core metric: a property's appearance rate (AR) is how often it shows up when AI recommends apartments. Higher AR, more visible. But AR on its own can fool you, in three ways:
- "Good" looks totally different from city to city. In one market, showing up 10% of the time is excellent. In another, you'd need 50% to stand out. So you can't compare a property in one market to a property in another without adjusting for that.
- Rank doesn't tell the whole story. A property ranked 30th but only 1% behind the leader is actually doing great. A property ranked 2nd but 10% behind the leader isn't as good as "2nd place" makes it sound.
- Even "how far behind #1" can mislead. Being 5 points behind a leader who scores 15% is a big gap. Being 5 points behind a leader who scores 50% is tiny. Same 5 points, very different meaning.
So we built one number that rolls all three together: MAAR.
MAAR tells you how visible you are compared to the best property in your own market. Score a 60% MAAR and you're 60% as visible as your local leader, whether that leader shows up 12% of the time or 45%.
What top performers have in common
Three patterns show up across the top performers:
- Narrower geographic focus. They concentrate in a handful of metros instead of scattering nationally. That focus produces deeper neighborhood and commute content, which AI models reward.
- Deeper property-website content. Real content programs at the community level: neighborhood guides, longer floor plan descriptions, price-and-amenity transparency. Not template-shipped pages. Property websites are now the most-cited source after the ILS network.
- Broader ILS distribution beyond Zillow. They list across the CoStar network and RentCafe. When the pipeline balance shifted toward CoStar this spring, they were already there.
What low performers have in common
The patterns at the bottom of the leaderboard are just as telling. Low-performing PMCs share a few traits:
- Thin ILS distribution, often only 2-3 platforms, usually all from the same data source.
- Template-heavy property websites with little unique content. Generic copy, stock photos, same floor plans every reader sees on every other site.
- Inconsistent listing data across platforms. Different pricing on Zillow vs Apartments.com. Missing unit details. Stale availability. AI picks up on the mismatch.
Takeaway: Visibility is earned at the community level, not the portfolio. The PMCs winning execute the fundamentals across every individual property: accurate listing data, strong property websites, broad ILS distribution. Portfolio size and budget don't decide it.
Your website vendor matters, but it's not the only lever
The platform your property website runs on is linked to how often AI engines cite you.
We've classified the website vendor for 1,793 unique competitor properties. To compare them fairly across very different markets, we rank by MAAR (Market-Adjusted Appearance Rate): how visible a property is versus the best one in its own market, scored 0 to 100. See Section 4 for how we built it.
MAAR by Vendor
Properties measured = unique competitor properties on each vendor with at least 2 AI appearances. We exclude single-appearance properties to cut noise.
Data as of May 28, 2026
Takeaway: Repli leads on MAAR, with Apartments247, Jonah Digital Agency, Entrata, and RentCafe close behind. Vendor choice matters, but the spread between the top vendors is small, and strong content moves the needle more than switching vendors does. A template alone won't do it: pair a strong vendor with content-rich listings (real photos, real prices, real amenity tags) to win on both fronts.
Ready to close the gap?
Pull a free Gemini report on any property, then talk to our team about closing the gaps across every engine.
A note on the data
Everything in this report comes from Peek's AI discovery dataset, December 4, 2025 to May 14, 2026: 795 assessments of subscriber properties against 3,811 top competitors. Each assessment runs renter-style prompts ("Best two-bedroom apartments near downtown Austin under $2,500") across ChatGPT, Claude, Gemini, and Perplexity, then captures which properties get cited and which sources the models pull from.
Comparisons labeled "Original Report" or "Mar 2026" come from the first Black Book. "May 2026" values are recomputed from a partial-May window (about 14 of 31 days), flagged with an asterisk where it could meaningfully change the reading.
AI Discovery Score is Peek's metric: the share of relevant AI queries where a property shows up. Think organic search ranking, but for AI prompts.