Read this first
This is the short, plain version of a long research paper. The full paper is called Borrowed Trust: The State of Native Advertising in 2026. It has the sources, the tables and the math. This page has the same findings in fewer and shorter words. The counts of firms, sources and checks 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 27 September 2026. It looks mostly at the US.
What native ads are
You have seen them. You read a news story, and under it sits a box of links. "Around the web." "You may like." Some of the links go to more stories on the same site. Some are ads. A native ad is an ad made to look like the stuff around it. It sits in the feed or the story, not in a banner at the side.
This research looks at the firms that fill those boxes and feeds on news and other sites. The two big ones are Taboola and Teads. Outbrain bought the old Teads in 2025 and took its name. Chapter 1 sets out what counts.
Ads in social feeds look like posts, so they are native too. So are paid spots in search. The paper treats those as rivals for the same money, not as part of this market.
What we found
1. This research found no size for this market
The only public number for "native" this research found is $147.98 billion. That is eMarketer's figure for US "native display" in 2026. But that figure counts ads in social feeds. In old forecasts that showed a split, social was 74% to 84% of the total. The last split came out in 2019. There is none for 2026. The ad trade's own count of what it earns has no line for native at all.
So how big is the part on news and other sites? No source this research found has measured it. The two big firms that trade on US stock markets took in $3.21 billion in 2025. That is about 1.6% of what Meta made from ads. But it is all their sales, around the world, and not all of it is native. So it is not the size of the market. Chapter 2 shows the math.
2. It is a business built on deals with sites
You would think the power sits with the brands that buy the ads. At Taboola, the biggest firm, it does not. Its ten biggest ad buyers bring in less than a tenth of its money. But two partners, Yahoo and Microsoft, bring in about a third of what it earns from ads on other firms' sites. Teads does not share these numbers, so this rests on Taboola alone.
To keep big sites, the firms sign long deals. They often promise a site a floor price for each thousand page views. In 2024, keeping those promises cost Taboola extra. The extra was about 18% of its bill for ad space, which is mostly money paid to sites. By mid-2026 it was about 13%. Chapter 3 and chapter 4 follow the money.
3. Many readers do not see them as ads
Tests have shown people native ads and asked what they saw. In three studies of paid stories and ads in feeds, only 7% to 37% of people saw them as ads. In one of those tests, 81% spotted a plain ad. The US trade watchdog, the FTC, ran a small lab test with 48 people, using search ads and native ads. About 47% saw the ads as ads. With the FTC's own fixes, 68% did. It was the only test that used a box of links. In these tests, good labels helped. But fewer people spotted native ads than plain ads.
These tests are old. They ran from 2014 to 2018, on some kinds of ads and some groups of people. They do not tell you how many readers miss the ads today. Chapter 5 has the studies.
4. If you pay per click, the risk stays with you
The help pages of Taboola, Outbrain and MGID describe ads sold by the click. You can tell the system that your goal is a sale. The system then picks which clicks to buy for you. But you still pay for each click, sale or no sale. So the risk stays with you. A target price per sale is a setting that steers the bids. It is not a promise to charge you per sale.
The big firms also sell by the view and by the sale. They do not say how much of their money comes from each. Chapter 6 shows how the system works.
5. The proof is missing, not bad
The key question for a buyer is simple. Did the ads cause sales that would not have come anyway? The best way to know is a holdout test. You keep the ads from one group of people, then you compare the two groups. This research looked for such a test on these widgets, with its design and results made public. In the sources it checked, it found none.
What the firms show instead are studies that set people who saw the ads next to people who did not. That sounds fair, but it is not. The people who see an ad are not like the rest. In big Facebook tests, studies built that way made the ads look about three to thirteen times as good as they were, at the midpoint. This research found no such number for native ads. So treat it as a warning, not a fix. Chapter 7 goes through it.
Fair tests of native ads in other places, on Facebook and in search, did find that the ads worked. The gap is in the box of links on news sites. So the right read is "not proven", not "proven not to work".
6. The big firms are moving past the widget
Both big firms now sell more than the box of links. Taboola sells display ads, video and ads inside AI answers, through tools called Realize and DeeperDive. At Teads, ads on TV sets that stream shows made up 13% of its sales in the second quarter of 2026. Neither firm says how much growth these new lines bring.
At the same time, some big sites report fewer visits. People Inc., a big US publisher, said its core site visits fell 22% in the second quarter of 2026 from a year before. It blames Google's AI answers, which show up before anyone clicks. Fewer visits likely means fewer spots for the widget. No one reports that number, so it is our read, not a fact on record. Chapter 11 tracks the signs.
7. On the big self-serve tools, you see where ads ran only after they run
On both big firms' self-serve tools, you get the list of sites where your ads ran once the ads start to run, not before. You can block sites. Their help pages show no list to pick from up front. You may get one in your contract. Teads' other buying tool, Outbrain DSP, does let you list the sites you want. A smaller firm, Readpeak, shows its list before the ads run. So it can be done. Chapter 10 covers what buyers can see.
8. AI buying agents go through the firms' own doors
New tools let AI agents buy ads for people. Most still ask a person to say yes first. Two groups are writing the rules for how those agents talk to sellers. On the pages this research checked, neither lists a big native firm as one that has built to its rules. Taboola runs its own server that lets AI agents read and change ad campaigns on Taboola instead. Chapter 11 has the detail.
9. One firm is out front, and no firm scores high on tools to check results
The paper scores 24 firms on 12 areas, from how they bid to how they help you measure. A score counts only what a firm shows in public. It does not measure how well its ads work. The scores then roll up into one index, the Native Quotient.
Only 9 firms had enough on record to place in a band. Taboola is the only one in the top band. 6 firms are in the next band and 2 in the one below that. The other 15 had too little on record to band at all.
On tools that help a buyer check results, no firm scores above 3 out of 5. No other area has a lower top score.
Read the bands, not the order. Small changes in how the index is built move firms up or down a few places. Even the bands can shift. Only 4 of the 9 firms keep their band under every test. Chapter 9 has the scores and the index.
If you buy these ads
Three things to do:
- Treat the sales a firm reports as a claim, not a fact. Ask how long after a click or a view a sale still counts. Ask how long sales take to show up, and what counts as a sale. It can be just a page view.
- Do the math before you test. A fair test needs a lot of people. Here is an example. Say 1% of people buy, and you want to spot a rise to 1.1%. With the usual test settings, you need about 163,000 people in each group. That is people, not clicks. Your own numbers will change this. If the math works, run your own holdout test, and expect a rough answer. If your test is too small to spot a change that matters to you, say so. Cap what you spend on ads you can't measure.
- Ask for the site list before the campaign and the full log of where the ads ran after it. Let the answer shape the price you pay.
Chapter 10 has the full advice and the math.
If you run a site
These deals pay, and the money can pull against the quality of your pages. In 2018, some sites gave up a promised minimum for a plain share of the money. They wanted more say over where the widget goes. Since then, promised minimums have come back at Taboola. Chapter 4 has the cases.
How this was made
This research was done with AI help and a public method. AI agents opened 977 sources and logged each one. Some could only be read in part. A second AI agent checked the key claims, with orders to try to prove them wrong. 317 of the 642 claims have now been checked against their sources. The rest were not checked a second time. Two outside reviews pushed back on drafts. Each change, 277 in all, is in a public log. The research has not been audited or peer reviewed. No firm in it saw it before it went out. Chapter 12 has the method.
The author, Evgeny Popov, has worked at Samba TV since May 2025. Samba TV measures TV viewing and sells ad data. Before that he worked at Verve Group, Hearts & Science and Lotame. None of them is scored. He co-leads a work group at the body that writes AdCP, one of the rule books for AI buying agents. Samba TV is a member of that body, and Teads names Samba TV as a partner for measuring ads. Teads is scored in the paper. Teads also sells ads on TV, which overlaps with Samba TV's market. No one paid for this research.
Read the full paper
The full paper has the sources, the data and each step of the math. Read Borrowed Trust in full. The data files are in its last chapter.