Opinions from people who say they used it: a tracker study
For four tracker brands, we counted the opinions a model labelled as from people who say they used the product. The label can be wrong, and it misses some people who do say it, so treat each share as an estimate.
Three findings
- Oura store reviews: the model labelled 726 of 781 opinions (93.0%), on Amazon, December 2024 to June 2026.
- Oura comments and replies: the model labelled 2,136 of 3,636 opinions (58.7%), on 4 platforms, November 2024 to August 2026.
- Amazfit post text: the model labelled 96 of 110 opinions (87.3%), all on Reddit, September 2025 to July 2026.
We measured four sleep and health tracker brands: Oura, Fitbit, Amazfit and RingConn. Every figure is per brand and gives its base. For a consumer insights team, the label sorts opinions by what they can tell you. It is not a test of honesty.
What share of opinions come from people who say they used it?
The table splits each brand’s opinions by who is speaking. Each cell gives the opinions the model labelled as from people who say they used it, of all opinions in that group.
| Voice | Oura | Fitbit | Amazfit | RingConn |
|---|---|---|---|---|
| All opinions | 4,294 of 6,438 (66.7%) | 375 of 797 (47.1%) | 456 of 745 (61.2%) | 393 of 507 (77.5%) |
| Store reviews | 726 of 781 (93.0%) | 6 of 7 | None | 169 of 185 (91.4%) |
| Spoken in videos | 1,264 of 1,725 (73.3%) | 102 of 280 (36.4%) | None | 79 of 97 (81.4%) |
| Post text | 168 of 296 (56.8%) | 57 of 110 (51.8%) | 96 of 110 (87.3%) | 13 of 19 |
| Comments and replies | 2,136 of 3,636 (58.7%) | 210 of 400 (52.5%) | 360 of 635 (56.7%) | 132 of 206 (64.1%) |
| Accounts with a labelled opinion | 1,454 of 2,269 | 231 of 422 | 208 of 308 | 89 of 136 |
Bases: Oura, 6,438 opinions on 6 platforms, December 2024 to August 2026. Fitbit, 797 opinions on 6 platforms, December 2022 to July 2026. Amazfit, 745 opinions, all on Reddit, September 2025 to August 2026. RingConn, 507 opinions on 6 platforms, May 2026 to August 2026.
How to read the table:
- Comments and replies include replies in forum threads, such as Reddit, as well as comments under videos and posts.
- Spoken in videos covers every speaker in a video, including people other than the creator. Post text is the caption or body of a post.
- Amazfit opinions are all on Reddit, so it has no store reviews or video opinions.
- Small groups get no percentage under 30 opinions. Fitbit’s 7 store-review opinions have no stored author name, so they sit in no account count. RingConn’s spoken opinions come from 6 accounts.
What share comes from people who say they used it on each platform?
A platform figure mixes voices. YouTube, for example, holds what is spoken in videos and the comments under them. Split by voice before you compare platforms.
| Platform | Oura | Fitbit | RingConn |
|---|---|---|---|
| 1,478 of 2,331 (63.4%) | 133 of 220 (60.5%) | 119 of 160 (74.4%) | |
| YouTube | 1,349 of 2,185 (61.7%) | 155 of 395 (39.2%) | 58 of 88 (65.9%) |
| TikTok | 557 of 839 (66.4%) | 55 of 112 (49.1%) | 17 of 28 |
| 177 of 265 (66.8%) | 23 of 54 (42.6%) | 0 of 5 | |
| None | None | 30 of 41 (73.2%) | |
| X | 7 of 37 (18.9%) | 3 of 9 | None |
Bases: Oura, 6,438 opinions on 6 platforms, December 2024 to August 2026. Fitbit, 797 opinions on 6 platforms, December 2022 to July 2026. RingConn, 507 opinions on 6 platforms, May 2026 to August 2026.
Amazon holds the store reviews, so its figures are the store-review row in the voice table. Amazfit’s opinions are all on Reddit, so its Reddit figure is its total.
The opinions in this study came from six products we chose to search. So platform figures describe our collection, not the web.
How should an insights team use this label?
Use it to sort opinions by the question they can answer.
- Split by voice. Store reviews, speech in videos, post text and comments answer different questions.
- Compare like with like. Compare brands within one voice or one platform, because the source mix moves a brand’s total.
- Count accounts as well as opinions. One speaker can give many opinions in one video.
- Write the base every time. Say “X of N opinions from people who say they used it, on P platforms, from month to month.”
- Treat each share as an estimate. The label gets some opinions wrong and misses some people who do say it. Both kinds of error may fall unevenly.
- Keep the other opinions. An opinion without the label is not proof that the person did not use the product.
- Label sponsored posts and count identical text once. Read sponsored posts apart from the rest.
- Read the quotes behind each label. Sample them by hand before you publish a number.
For the signs to look for in one post, read how to tell if a product opinion comes from someone who says they used it. To turn these opinions into answers before a survey, see our preliminary research guide.
How we measured
- Data. We used Honestly’s proprietary dataset of product opinions. Most sources are public; the record of each opinion, linked to its source, is ours.
- Scope. Six products: Oura Ring 4 and Ring 5, RingConn Gen 3, Fitbit Inspire 3 and Fitbit Air, and Amazfit Helio Strap. An opinion counts for a brand when the model tied it to one of that brand’s own products. Opinions about rival brands, or about other companies’ accessories or apps, are left out. Snapshot date: 14 November 2026.
- The label. Honestly labels opinions where the person says they used the product. The label is a model’s reading of the words, not proof of use or purchase. It can be wrong, and it misses some people who do say it.
- Audit. Before we published, we read a sample of labelled and unlabelled opinions for each brand by hand. In that sample, some labels were wrong, and some unlabelled opinions also said the person used the product.
- Screens. We set aside posts with a sponsorship disclosure we could detect, and opinions where a brand made claims about itself. Identical opinion text from different accounts counts once, which also merges short common phrases such as “I love it”.
- Dates. Each date range covers the middle 80% of a brand’s opinions, from the 10th to the 90th percentile of post dates. We left out post dates before 2000.
- Accounts. An account is one author name. The same name on two platforms counts once, and one person with two names counts twice.
- Per brand. We print no total across brands.
Frequently asked questions
Does “says they used it” mean the person bought the product?
No. It means the model read the words as saying the person used it. They may have borrowed it, been given it or tested a review unit.
How reliable is the label?
It is a model’s reading of the words, not proof of use or purchase. In our hand audit, it got some labels wrong and missed some people who do say they used it. So treat every share as an estimate, and read a sample of quotes before you publish a number.
Why is there no ranking of voices or brands?
The wrong labels and the misses may not fall evenly across voices, platforms or brands. A gap of a few points between two groups could close or flip. So read each share on its own, with its base.
Should we drop opinions without the label?
No. An opinion without the label is not proof that the person did not use the product. Keep it and read it as a different kind of evidence.
Why is every figure per brand?
In this study, Oura’s base of 6,438 opinions is the largest of the four brands. A total across brands would give Oura the most weight.
Bring one product question. See what the dataset already holds on it.
Request a demo
