On this page 10 sections
Explore the public leaderboard and the full scoring methodology.
We tested every real estate website we could find for agents in Metro Vancouver in September 2026. 1,395 of 6,510 sites — 21.4% — returned no usable home page to our automated requests. These sites belong to 2,009 agents.
The most common result, covering 790 of those agents, was an HTTP 403 "Forbidden" response. The next most common, covering 586, was a domain name that did not resolve. The rest were refused connections, timeouts, certificate errors, missing home pages and server errors. We did not check these sites in a normal browser, so we cannot say how they behave for human visitors.
Of the 17,507 agents in our data, 7,583 (43.3%) have a website that returned a readable page to our test. 1,906 agents reach our highest readiness band — a combined score of 70 or above out of 100, on the two measures described below. That is 10.9% of all agents, or 25.1% of the agents whose site we could read.
What can our automated website test read?
When a buyer asks ChatGPT, Claude, Perplexity or Google's AI Overviews "who is a good real estate agent in Coquitlam," the answer may cite web pages. We tested whether our automated requests could fetch an agent's website and find text clearly linked to that agent. A failed request in our test does not prove that every AI product cannot access the site: in our small Perplexity sample, one site we could not fetch was still cited. An engine may also name an agent from other sources, such as directories or reviews.
This study measures three problems our automated test can encounter:
- The request fails. The domain does not resolve, the server refuses the connection or times out, the TLS certificate is broken, or the server answers with an error such as 403 Forbidden.
- The page arrives empty. Many crawlers do not run JavaScript. Google Search renders JavaScript, although our test did not. A site that builds its text in the browser can give a crawler that does not run scripts an almost empty page. In our sample, 20.3% of reachable sites had fewer than 400 characters of text in the raw HTML.
- The page has few clear facts. The page loads, but it gives few specific facts linked to a named person, such as years of experience or the neighbourhoods they serve.
Search engine optimization still matters. Google says the same basic SEO practices apply to AI Overviews and AI Mode as to Google Search in general. It also says no special files, markup or schema.org structured data are needed to appear in them. Our content score adds one more question: does the page state clear facts about the agent that an engine could quote?
How many Metro Vancouver agents have a website our test could read?
Of the 17,507 agents in our data:
| Result | Combined score | Agents | Share |
|---|---|---|---|
| Highest band | 70–100 | 1,906 | 10.9% |
| Middle band | 50–69.5 | 4,099 | 23.4% |
| Low band | 0–49.5 | 1,578 | 9.0% |
| Site returned no usable page | — | 2,009 | 11.5% |
| No website found | — | 7,915 | 45.2% |
9,924 agents — 56.7% — have no website that returned a page to our test: either we found no website, or the website we found did not return a usable page.
We found a website for 9,592 agents. 2,009 of them — about one in five — have a site that returned no usable page to our automated requests.
Counted by website rather than by agent, the median combined score was 63.5 out of 100 (n=5,115). The median content score was 48 out of 100 (n=5,033).
What did the websites of 2,009 agents return?
These 2,009 agents shared 1,395 distinct websites that returned no usable home page. We requested each site over HTTPS, then over HTTP if that failed. The table counts affected agents, so a shared site can appear more than once. The last column lists common causes of each result. We did not confirm the cause for any single site.
| Result we observed | Agents | Share | Common causes |
|---|---|---|---|
| HTTP 403 Forbidden | 790 | 39.3% | Bot protection or firewall rules, access rules on the server |
| Domain name did not resolve | 586 | 29.2% | Expired registration, missing DNS records |
| Connection refused | 243 | 12.1% | Hosting stopped, server not listening |
| Timed out (8 seconds to connect, 15 seconds in total) | 231 | 11.5% | Slow or overloaded server, network filtering |
| TLS certificate error | 80 | 4.0% | Expired or misconfigured certificate |
| Home page returned 404 | 38 | 1.9% | Missing page or routing error |
| Other errors (5xx, 429 and others) | 41 | 2.0% | Server faults, rate limiting |
Timeouts, 429 responses and 5xx errors can be temporary. Each result above records our test in the week of 21 September. It does not prove the site is down for all visitors or down permanently.
Re-test with a browser user agent. We tested the sites in two batches. In the first batch, 451 domains returned a 403, a timeout, a refused connection, a 429 or certain 5xx errors. We requested each of them again, this time with a Chrome browser user-agent string. 40 of the 451 (8.9%) returned a page. All 40 had first returned a 403. We counted these 40 as reachable and scored them. The other 411 failed again. We did not re-test the 642 failing domains from the second batch, including 230 that returned a 403.
This re-test changed only the user-agent string. It did not run a real browser. So it cannot show whether the remaining sites open normally for people.
Failures cluster on shared domains
- Of the 790 agents whose site returned a 403, 261 share a single brokerage domain.
- Of the 243 agents whose connection was refused, 100 share a single brokerage domain. A further 38 agents from the same brokerage had a domain that did not resolve.
When many agents share one domain, a single server or DNS setting decides the result for all of them. The websites of 908 affected agents (45.2% of 2,009) returned a 403, a certificate error or a missing home page. These results usually come from how a server is set up, not from the site's content. We did not confirm the cause, or whether a fix would be simple, for any individual site.
How many sites block AI crawlers?
Few. Of the 7,583 agents whose site returned a readable page, 502 — 6.6% — have a site whose robots.txt file blocks at least one of the six AI crawlers we checked. Counted by website, it is 419 of 5,115 sites (8.2%).
| Crawler blocked | Agents |
|---|---|
| GPTBot (OpenAI) | 476 |
| ClaudeBot (Anthropic) | 279 |
| Google-Extended | 249 |
| CCBot (Common Crawl) | 49 |
| PerplexityBot | 38 |
| OAI-SearchBot | 38 |
We cannot tell from the data whether a block was the agent's own choice. One pattern suggests that many were not set site by site: 158 of the 419 blocking sites block exactly the same three crawlers (GPTBot, ClaudeBot and Google-Extended). 157 of those 158 also have the same structured-data setup, which points to a shared website platform or template. We did not confirm this with any platform.
What each block does depends on the product. Google says Google-Extended does not affect whether a site is included or ranked in Google Search. It controls whether Google can use the content to train Gemini models and to ground answers in some other Google products. GPTBot and ClaudeBot mainly collect content for model training. OpenAI uses a separate crawler, OAI-SearchBot, for search results in ChatGPT.
Our technical score gives points for allowing all six crawlers. A site that blocks GPTBot, ClaudeBot and Google-Extended loses 15 of 100 technical points. For example, one site scored 81 on technical access and 70 on content. On our scale, removing its three blocks would raise its technical score from 81 to 96. That is a change in our score. It does not show a change in how Google Search treats the site.
Overall, 93.4% of agents with a readable site have a site that allows all six crawlers we checked (91.8% counted by website). For most agents, crawler rules are not what keeps their site out of AI answers.
What do the 100 publicly ranked websites have in common?
Our public leaderboard selects the 100 highest-scoring eligible websites for individual agents. Shared team and brokerage websites, and sites where we could not confirm the agent's name, are excluded. These eligibility rules make the published top 100 a selected group, not a representative sample of all agents. The sites name 58 different verified brokerages; three do not state a brokerage.
This table counts websites. "All scored websites" means the 5,033 sites that returned a usable page and received a content score.
| Practice | Public top 100 | All scored websites |
|---|---|---|
| Has question-style subheadings | 93% | 43.2% |
| States name, role and city in one passage | 86% | 32.7% |
| Uses FAQPage structured markup | 27% | 1.1% |
These practices are part of our content score. The table shows what high-scoring sites contain; it does not show that these practices cause AI citations.
| Websites | Technical score (mean) | Content score (mean) |
|---|---|---|
| Public top 20 | 95.6 | 82.1 |
| Public top 100 | 94.8 | 73.2 |
| All reachable websites | 70.3 (n=5,115) | 42.8 (n=5,033 scored) |
Content is the weaker area across the market: the median technical score is 77, but the median content score is 48. Only 17 of 5,033 sites scored 80 or above on content.
Which areas score best?
Port Coquitlam, New Westminster and Pitt Meadows have the highest share of agents in our highest band. Vancouver and Richmond have the lowest.
Areas are the service areas listed in our agent data. There are 20 of them. They are not the same as municipal boundaries. For example, the Vancouver row includes 1,199 agents whose offices are in North Vancouver or West Vancouver. 1,970 agents (11.3%) list more than one area — 3.26 on average, and up to 16 — so they appear in several rows. The rows shown sum to 21,703 rather than 17,507. Only areas with 150 or more agents are shown; 13 agents list only smaller areas.
| Area | Agents | Highest band | No website found |
|---|---|---|---|
| Port Coquitlam | 189 | 24.9% | 23.8% |
| New Westminster | 354 | 22.9% | 38.4% |
| Pitt Meadows | 233 | 20.6% | 25.3% |
| Port Moody | 402 | 19.7% | 29.4% |
| Langley | 1,021 | 19.5% | 15.1% |
| White Rock | 264 | 19.3% | 30.3% |
| Coquitlam | 1,337 | 16.5% | 37.8% |
| Maple Ridge | 384 | 15.9% | 30.2% |
| Delta | 437 | 13.0% | 35.9% |
| Surrey | 3,635 | 11.7% | 22.1% |
| Burnaby | 1,799 | 11.0% | 58.8% |
| Vancouver | 9,324 | 10.9% | 51.5% |
| Richmond | 2,324 | 7.4% | 68.4% |
In Richmond, we found no website for 68.4% of its 2,324 agents, and 173 agents are in the highest band. Burnaby and Vancouver are also above 50% with no website found.
We did not measure why. One possible reason is channel choice: some agents in these areas work mainly through WeChat, Xiaohongshu and referrals, where a public website plays a smaller part. Another is our own coverage: our search may miss more sites in these areas, for example sites under a Chinese-language brand name. A low share in the highest band means we could not read a strong website for those agents. It says nothing about how good their business is.
Are higher-scoring sites cited more often by AI?
We did a small test. We put 12 buyer-style questions to Perplexity — covering different areas, property types and buyer situations — and recorded every domain it cited. Then we looked up those domains in our own scores.
The 12 questions produced 51 distinct cited domains. 27 of them were in our dataset. We set aside 6 of those 27 because they are not individual agent or team sites: two property portals (rew.ca, zolo.ca), two brokerage sites (remax.ca, multiplerealty.com), a condo project listing site (vancouvernewcondos.com) and a presale marketing company (bridgewellgroup.ca). That left 21 agent or team sites.
This comparison counts websites, not agents, so each domain counts once. That is why the market figures here differ from the agent-level table above (71.3 and 44.4).
| Websites | Technical (mean) | Content (mean) | Combined (mean) |
|---|---|---|---|
| Agent and team sites Perplexity cited (n=21) | 84.6 | 54.0 | 69.3 |
| All reachable sites | 70.3 (n=5,115) | 42.8 (n=5,033) | 56.2 (n=5,115) |
82 reachable sites could not be scored for content. They count as 0 for content in the combined score. Two of the 21 cited sites are affected by gaps in our test: one returned no usable page, and one returned a page but could not be scored for content. Both are included, with 0 for the parts we could not measure.
Against all reachable sites, the cited sites had a median percentile of 91. 11 of the 21 were in the top 10%.
In this small test, cited sites scored higher on average. But the test is small: 12 questions, one engine, one point in time. We did not control for other factors, such as how well known an agent already is. The two cited sites in the bottom quarter are the two our test could not fully measure. Perplexity could still cite them, so our test misses some sites that AI engines can use. We only looked at sites that were cited, so this test cannot tell us how often high-scoring sites are not cited.
Directories appear often
Many cited domains were not agents' own sites. In 9 of the 12 answers, Perplexity cited at least one agent directory or property portal. The most frequent were agentpronto.com (6 answers), rankmyagent.com (5), zillow.com (4), rew.ca (3) and rate-my-agent.com (3).
This suggests that a complete profile on a major directory may also matter for AI answers. We did not measure how much. A directory profile and your own website do different jobs: on your own site, you control every fact that is written about you.
What can an agent check first?
This list is our suggestion. It is based on how our score works and on how rare each practice is in our data. We did not measure how much any single change affects AI citations.
| # | Check | Time | What our data shows |
|---|---|---|---|
| 1 | Confirm your site answers a request from a non-browser client, such as curl | 15 min | 11.5% of all agents had a site that returned no usable page to our test |
| 2 | Review which AI crawlers your robots.txt blocks, and keep only the blocks you want | 15 min | 6.6% of agents with a readable site block at least one; some blocks come from a template, not a choice |
| 3 | Add RealEstateAgent JSON-LD with name, phone, service area and brokerage | 1 hour | 40.2% of sites carry it |
| 4 | Write one sentence naming yourself, your role and your city together | 10 min | 32.7% of sites do this |
| 5 | Put client reviews on the page as text; add Review markup if you use structured data | 1 hour | 43.6% of sites have review text; 1.9% mark it up |
| 6 | Add a short FAQ section; FAQPage markup is optional | 2 hours | 1.1% of sites have FAQPage markup |
| 7 | State your years of experience and name your neighbourhoods | 30 min | 77.4% do not state experience; 55.6% name fewer than three areas |
| 8 | Delete generic marketing phrases | 30 min | 52.6% of sites use them |
Items 5 and 6 are the rarest in our data. Across 5,033 sites, 98 carry review markup and 55 carry FAQPage markup. Markup is not required: Google says AI Overviews and AI Mode need no special structured data. What matters more is the text itself — real reviews and clear questions with direct answers give an engine specific facts about you.
The phrases in item 8 look like this: your trusted partner, passionate about real estate, going above and beyond, second to none. None of them states a fact that anyone can check.
Frequently asked questions
How many real estate agents are there in Metro Vancouver?
We identified 17,507 agents in Metro Vancouver in September 2026. By listed service area, 9,324 list Vancouver (including agents with offices in North and West Vancouver), 3,635 Surrey, 2,324 Richmond and 1,799 Burnaby. Some agents list several areas.
What percentage of real estate agents have a website?
We found a website for 9,592 of 17,507 Metro Vancouver agents, or 54.8%. Of those, 2,009 had a website that returned no usable page to our automated requests. The true share with a website is likely a little higher, because our search does not find every site.
Do AI search engines read JavaScript websites?
It depends on the crawler. Google Search renders JavaScript. Many other crawlers only read the raw HTML. In our sample, 20.3% of reachable real estate sites had fewer than 400 characters of text in the raw HTML. A crawler that does not run JavaScript sees an almost empty page on those sites.
Which AI crawlers should a real estate website allow?
Our score checks six: GPTBot and OAI-SearchBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot, Google-Extended and CCBot (Common Crawl). Each one controls something different. For example, Google says Google-Extended does not affect inclusion or ranking in Google Search. OpenAI uses OAI-SearchBot for ChatGPT search, and GPTBot for model training. Check each company's documentation before you decide what to block.
What is FAQPage schema and why does it matter for AI?
FAQPage is a structured data format that marks a question and its answer as a pair. It is not required for AI features. Google says AI Overviews and AI Mode need no special structured data. The useful part is the clear question-and-answer text itself. Only 1.1% of the 5,033 content-scored websites use FAQPage markup.
Does a good website guarantee an AI engine will recommend me?
No. In our small test of 12 buyer questions on Perplexity, the 21 cited agent and team sites had an average combined score of 69.3 out of 100, against 56.2 for all reachable sites. But the test is small, and it cannot show how often high-scoring sites are not cited. AI engines also draw on directories, reviews and press coverage.
Is a low AI visibility score a sign of a bad agent?
No. The score measures only how well our test could read a website and find clear facts in it. It says nothing about experience, service or sales results. Some agents work mainly through referrals, WeChat or Xiaohongshu, where a website plays a smaller part.
How often does AI visibility change?
It can change whenever a website, its hosting or its robots.txt changes. An expired domain or a new server rule can stop a site from answering automated requests within a day. The agent may not notice, because the change may not affect what they see in their own browser.
How this study was conducted
Who. Maahes AI, an AI visibility consultancy based in Vancouver, British Columbia. We are not affiliated with any real estate board, brokerage or regulator.
What. 6,510 distinct real estate websites belonging to agents in 20 service areas across Metro Vancouver. 5,115 returned a usable home page; 5,033 of those were also scored for content.
When. 21 to 24 September 2026.
How. Each site received a small number of rate-limited requests that identified our crawler: the home page, robots.txt, sitemap.xml, llms.txt, and an About page where one could be found. We did not log in or submit forms. In the first of two batches, 451 failing domains were requested again with a browser user-agent string. We did not open any site in a real browser.
Scoring. Two separate scores, each out of 100. The combined score is their exact average, shown to one decimal place. Readiness bands: highest 70 or above, middle 50 to 69.5, low below 50. Sites that returned no usable page are reported separately, not scored.
| Technical accessibility | Points | Content readiness | Points |
|---|---|---|---|
| AI crawler access (six crawlers) | 30 | Structure (headings, lists, short paragraphs, question headings) | 25 |
| Text in the server-rendered HTML | 25 | Extractable facts (numbers, years of experience, neighbourhoods, specialties, FAQ markup) | 30 |
| Structured data | 20 | Entity clarity (name, role and city together; brokerage; credentials) | 20 |
| Name and phone in the HTML | 10 | Authority signals (outside links, reviews, quotes) | 15 |
| Discoverability (title, description, sitemap, llms.txt) | 15 | Plain language (10 points, minus 3 for each generic marketing phrase) | 10 |
| Total | 100 | Total | 100 |
A page marked noindex loses 30 technical points. The table summarizes the scoring criteria used for this report. Of the 5,115 reachable sites, 82 could not be scored for content; they count as 0 for content in the combined score. The published rankings use these scores without rounding the combined score before sorting.
Limits of this study
We measure readability, not citation. Whether an AI engine names an agent also depends on reviews, directories and press we did not measure.
The score includes choices that do not affect search. Our technical score gives points for allowing all six AI crawlers, including crawlers used for model training. An agent who blocks these on purpose loses points, even though, for example, blocking Google-Extended does not affect Google Search.
Home pages and About pages only. Sites with a lot of blog or neighbourhood content may score lower here than their full site deserves.
One moment in time. A site that was slow or mid-migration that week is recorded that way.
Coverage is incomplete. We could not find a website for 45.2% of agents. In a random sample of 250 agents without a website, a web search found a site for 10 (4.0%). At that rate, roughly 300 more agents have a website that a search could find, and a search does not find every site. The 45.2% figure is an upper bound.
Weights are a judgment. The equal split between technical and content reflects our reading of current AI engine behaviour. Both component scores are reported separately so readers can weigh them differently.