We installed Reveal on two of our own websites. Here is everything that happened.
Before asking anyone to trust Reveal, we put it on two websites we own and read every number. One sells software to businesses; the other is a landscaping company. These are the results, including the mistake Reveal made and how we fixed it.
The two websites
cguardpro.com
CGuard Pro: security-guard management software, sold to businesses (B2B).
Read after the first ~8 hours
innovationgrounds.com
Innovation Grounds: a landscaping company in Austin, Texas.
Read after the first ~3 hours
Both companies belong to Prynt's owner, which is why we can publish their numbers. Every figure here is an aggregate count; no visitor's personal data appears in this report.
Method
- We added the standard Reveal snippet to every page of both sites and pointed it at our staging instance, which runs the same code as production.
- We did not filter, tune or exclude anything before reading the numbers.
- We counted every session Reveal recorded and how it classified it: bot, home or mobile network, hosting or cloud network, or company.
- We checked the one company match by hand.
Results
| cguardpro.com | innovationgrounds.com | |
|---|---|---|
| Observation window | ~8 hours | ~3 hours |
| Sessions | 60 | 60 |
| Bots filtered | 11 | — |
| Home and mobile networks (anonymous, no credit used) | 39 | 4 |
| Hosting and cloud networks | 20 | 55, from servers in one Singapore data center that execute JavaScript (most likely monitoring or crawling) |
| Countries | 17 | — |
| Flagged as a company | 1 (a false positive, see below) | 0 |
— not counted separately in this window.
- Home and mobile networks
- Hosting, cloud and data-center networks
- Flagged as a company
- Other, no company
cguardpro.com: 39 + 20 + 1 = 60. innovationgrounds.com: 4 + 55 + 1 session that matched no company = 60. The bot filter's count (11 on cguardpro.com) is in Table 1 and is not broken out here.
The one company match was wrong
In the first hours on cguardpro.com, Reveal flagged one network as a company: Velocom S.A. de C.V., in Mexico. The name carries a corporate suffix and nothing in the data we used then marked the network as an internet provider, so the rules treated it as a business.
It is not a business network. Velocom is a consumer ISP. APNIC Labs estimates that about 216,000 people use it to get online. Naming that visitor “Velocom” would have been a false positive, exactly the kind of guess Reveal exists to avoid.
What we changed, and in what order
4–5 November 2026
- 01 · Install
The standard snippet goes live on cguardpro.com and innovationgrounds.com.
- 02 · First read
One company match out of 60 sessions on cguardpro.com: Velocom S.A. de C.V. (Mexico).
- 03 · Check
Checked by hand: a consumer ISP with ~216,000 users according to APNIC Labs.
- 04 · Fix
APNIC Labs end-user estimates join the network classification. A network with 1,000 or more measured end users is treated as an ISP, unless it is a known corporate network.
- 05 · Re-test
The public benchmark of 1,357 addresses runs again with the new rule: 99.7 % precision and 98.6 % coverage at “likely or better”, 100 % precision at “verified company”.
About those benchmark numbers
They come from a public test set of 1,357 IP addresses whose ownership can be checked in public records: 691 company networks and 666 that are not (home ISPs, mobile carriers, cloud ranges, VPN and Tor exits, crawlers, coworking operators). Coverage there means the share of the company networks in the set that Reveal named. Both figures are upper bounds: a curated public set is easier than real traffic, and they say nothing about how many of your visitors Reveal can name. On real traffic most visitors are not on a company network at all, as this test shows.
Most “traffic” is not people
That is the lesson we would give anyone evaluating a visitor-identification tool: most traffic is not people, and most people are not on a company network. A tool that shows you a company for every visit is guessing.
What this means for you
Where Reveal shines
- B2B websites whose buyers browse from the office: manufacturers, logistics, professional services, software sold to companies.
- Agencies running several client sites, who need evidence they can show a client.
- Teams in Europe and Latin America, where identifying anonymous people is not an option. Company-level plus your own first-party people is the model that works there.
When to pair it with first-party data
- Consumer or local-service sites, like a landscaping company: most visitors browse from home or a phone, so expect few companies. Use forms, booking links and logins; Reveal links those people to their sessions, with consent.
- Remote buyers: someone working from home looks like a home connection. Personal links in your emails and a short form bring them back honestly.
What we changed because of this test
- Network classification now uses APNIC Labs end-user estimates: a network that serves 1,000 or more people is an ISP unless it is a known corporate network.
- We re-ran the full public benchmark with the new rule and published the result above with its limits.
- Nothing else needed changing: home, mobile, hosting and bot traffic was classified as anonymous and cost no credits.
Limits of this test
- Two websites, a few hours each. This is a field check, not a statistical study.
- Our own sites get little office traffic. A B2B site with more visitors from offices will show more companies.
- The benchmark figures are upper bounds on a public test set, not a prediction for your traffic.
Run the same test on your site
The free plan covers one website and 100 companies a month. Install the snippet, wait a day and read your own numbers, including the anonymous ones.