Read this first
This is the short form of a long research paper. The full paper is called Last Seen: The State of Location Advertising and Intelligence in 2026. It has the sources, the tables and the math. This page has the same findings in fewer and shorter words. The counts come from the same data file as the paper. The other numbers are copied from it. Each part links to the place in the paper that backs it up.
The research runs to 24 October 2026. That is when facts that can change, like who owns a firm, were checked. It looks mostly at the US.
What location data is
Your phone knows where it is. Lots of apps ask to use that. Some of them pass it on to firms that buy it. Those firms sell it to those who buy ads, and to those who study shops, malls and parks.
The data does three jobs in ads. It builds groups of people to aim ads at, like "people who went to a gym this month". It counts store trips after an ad ran. And it helps price billboards and screens.
The paper keeps four kinds of number apart:
- Seen: a phone in the group was seen in the store.
- Scaled up: a model grew the seen count to guess the whole town.
- Told: the count came from sales, door counts or log-ins.
- Guessed from the ad: the place came from the ad call or the phone's internet (IP) address.
Every firm we checked that says what kind of count it sells sells the second kind. Most do not say. Chapter 1 sets out the terms.
What we found
1. Most visit counts are guesses, or don't say what they are
We looked at the firms that count store trips after an ad. Four say they sell a scaled-up count. Some sell a seen count too. Nine counts do not say what kind they are. So a chart that says "visits" may well be a scaled-up guess, and the firm may not tell you. Chapter 1 has the details.
2. No one shows how the phone groups grew or shrank
Apple and Google made phones ask more before they share place, mostly from 2019 to 2021. Many think that shrank the data. It may have. But in the sources we checked, no firm shows the size of its phone group the same way before 2021 and after 2023. And no filing puts a number on the loss. So we can't say. The firms also changed hands. T-Mobile bought Vistar Media and Blis in 2025. GroundTruth and Foursquare got new owners in 2026. Chapter 2 has the deals. Chapter 3 has the phone changes.
3. No one has checked the counts against real store traffic
The fair test is simple. Take a firm's counts for lots of stores. Put them next to door counts or till data from the same stores. See how far apart they are.
We found no such test by an outsider for any named firm. The checks we did find look at parks, trails, polls and the census. Most use data from before 2021. One found that an old phone group missed older and non-white voters. At one trail, a firm's count was 30.5% above the trail counters, though the two counted different areas and activities. Chapter 4 goes through the studies.
4. "Lift" is mostly a match, not a test
Lift is the extra visits an ad caused. The clean way to find it is to keep some people from seeing the ad, on purpose, and compare. That is a holdout.
Most firms do not do that as a rule. They compare those who saw the ad with those who look like them. That shows a link, not a cause. Nine firms say how they measure visits, and eight of them claim lift. Six of those eight use a look-alike group, and two use a holdout. The ninth counts visits and claims no lift. The group you compare with can change the answer a lot. In one firm's case, the same campaign showed 29% lift against its holdout and 310% against people who never saw it. Chapter 5 shows how lift is made.
5. Few firms say where their target groups come from
Firms sell groups of people based on where they went. Some data comes from apps. Some comes from the stream of ad calls. Most firms do not say which source is the main one. So we can't tell which kind leads. Phone carrier data is new here. Since May 2025, T-Mobile has sold its data through Vistar's screen-buying tool. On 5 October 2026 it added place data, sold only there. Chapter 6 has the details.
6. The law got tougher, but the trade goes on
The FTC polices trade in the US. It has five orders against firms that sold place data. Only one, on InMarket, bans the sale of exact place data. The other four ban the sale of trips to places like clinics and churches. Three of the five firms still say they sell exact or phone-level data. That does not mean they broke the orders. It means the orders did not end the trade.
Five states now ban the sale of exact place data. They are Maryland, Oregon, New Jersey, Virginia and Connecticut. Four of the bans began in 2026. One firm, Gravy, was hacked in January 2025. Chapter 7 has the record.
7. No one has signed off on a visit count
The MRC is the body that checks how ad numbers are made. Its only rules for place data date from 2017, and they say nothing about lift. In the past it passed two firms' visit data: Foursquare's in 2020 and GroundTruth's in 2021. Both have since lost it. On 24 October 2026 no visit count was on any of its lists. Chapter 8 has the details.
8. Billboards run on phone data too
Screens and billboards can't count who walks past. So the trade makes a guess from phone data and road traffic. Half the screen sellers we checked do not name the phone data they use. The US group that sets billboard counts is changing how it works, with a test in late 2026. Chapter 9 has the details.
9. AI buyers can describe a place, but no proof yet that they buy by it
New AI tools can buy ads by talking to other tools. Their shared rules can name countries, states, store areas and screens. But just one working tool sends place targets through, and only for big areas like states and TV markets. No public test shows it working from end to end. The author helps run the group that writes one of these sets of rules. His own plan for map cells is not in it yet. Chapter 10 has the details.
How the firms scored
We scored 47 products on 11 things a buyer needs to know. Each got 0 to 5, based only on what the firm shows in public. We put them in five groups, so like is only matched with like. The weakest score in each group was for outside checks. No firm has an outside body sign off on its counts. The checks against real store traffic were weak too. 1 product shows the most in public. That is Placer.ai, and only just. Its counts for a whole place are scaled-up guesses, though it shows the raw counts too. One rule is ours: when an FTC order, court finding or breach in the last three years is about a product's own location data, that product gets at most 1 point on privacy. We show its score without that rule too. A high score means a firm shows more of its work. It does not mean its numbers are right. Chapter 11 has the scores.
What to ask before you buy
Ask what kind of count it is: seen, scaled up, told or guessed. Ask if there was a holdout. Ask if anyone checked the counts against real store traffic. Ask where the data comes from, and how people said yes to share it. Ask if the firm is under an order. Chapter 12 turns this into eight questions.
How this was made
AI agents did the research from one set of rules. A second AI checked each file and tried to prove it wrong. A model from a different AI firm then read the work twice. Each fact it raised was checked at the source first. Some of its own points were wrong. All 493 facts the research found were checked.
The author works at Samba, a TV data firm. He helps lead a work group at AgenticAdvertising.org. He once worked at Verve Group. Verve once owned a firm that sold ads by place. No firm paid for, read or saw the paper first. AI helped make this work, and the method is public. It is not peer reviewed. Chapter 13 has the method.