Fortell Hearing Aids: What Is Actually Known So Far

Fortell is a New York hearing aid company that launched in December 2025 and has since raised $163 million at a $740 million valuation, been profiled by Wired and the New York Times, and named to Forbes' Next Billion-Dollar Startups. Its founders came from Butterfly Network. The device costs $6,800 a pair and is sold only at Fortell's own clinics, fitted by audiologists Fortell employs.

Why this concerns Westchester now. Most coverage frames Fortell as something arriving later, in cities that are still waitlists. That is not the position here. Two of its three open clinics are at 530 Park Avenue and on Greenwich Avenue, which puts one a short drive east of Harrison and the other a train ride south. Patients in this county can be fitted this month if they choose to be, with no waitlist involved. We would rather they went into that decision informed than discover the terms afterwards.

We should also declare our position, because you should weigh what follows accordingly. Fortell does not supply independent practices and has no intention of doing so, which makes this a competitor writing about a competitor. We have tried to write it the way we would want somebody to write about us.

The idea. Nearly every hearing aid advertising artificial intelligence works by classifying the room and adjusting to suit: this is a restaurant, so narrow the microphones and pull the noise down. Fortell's position is that classification is the wrong level to work at, and that the device should instead separate the individual sound sources and rebuild the mixture with the voices you want brought forward. That is source separation, the technique behind the noise suppression on a video call, and the reason it has not appeared in hearing aids before is delay. An aid that lags the room makes your own voice sound wrong and stops lips matching sound. Fortell built its own chip to get the delay to roughly ten milliseconds, which is inside the window where most wearers stop noticing. Whatever one makes of the marketing, this is a genuine engineering position rather than last year's processing renamed.

Following the evidence. The homepage claims the world's most intelligent hearing aid, up to 98 per cent of background noise eliminated, and ten times more words understood. Behind those sit two studies. The first, from November 2025, is a pre-registered, double-blind randomised trial of speech intelligibility in thirty hearing aid users, posted as a PDF on Fortell's own site. The second, dated May 2026, is a blinded preference study in twenty adults, and it is published: Spatial AI consistently preferred to state-of-the-art hearing aids in multitalker noise, International Journal of Audiology, and you can read it at https://doi.org/10.1080/14992027.2026.2663345. An earlier version of this page said we could not find that paper indexed outside Fortell's own site, and that neither study stated a sample size. Both statements were wrong. The corrections are set out at the foot of this page.

What we would still put to a reader is narrower than what we first wrote. Every author on the published paper is a Fortell founder, employee or compensated consultant, the study was funded by Fortell, and the paper declares as much in its statement of interest, which is more than a good deal of industry research does. Neither study names the make and model of what it was compared against: the preference study describes its five controls as the top-spec device from each major manufacturer, and the intelligibility study withholds its single control altogether. The January 2026 reporting on that trial named the control as Phonak's Audéo Infinio Sphere.

Two numbers need separating, because we ran them together. The 9.2 dB quoted in January and the 17 dB on the site today are different measurements from the same trial, and neither has moved. The first is the SNR-50 improvement, the shift in the noise level at which listeners get half the words right, measured against the control aid. The second is a laboratory measurement of signal-to-noise improvement over the unaided ear, where the control aid managed 1.9 dB. Our description of the headline figure as having changed was wrong.

The ten-times claim needs the same care. It is an odds ratio rather than a count of words. In the hardest condition tested, minus 6 dB, the model puts the odds of identifying a word correctly 18.9 times higher with the Fortell device than with the control, which the company converts to roughly ten times the probability for a typical listener; the raw group averages there are 50 per cent against 7 per cent. At 0 dB they are 80 per cent against 31 per cent. We set the 0 dB figures against a claim the company makes about the hardest condition, called the ratio two and a half, and have withdrawn it. What is left is a point about presentation and not about fact: a shopper reads ten times more words as ten times the words, and the number behind it is a modelled odds ratio at the single most difficult signal-to-noise ratio in the study. Ask about it.

One thing has not changed. The NYU researcher quoted on the homepage saying he had never seen so categorical a result also told the same reporter that the configuration used was particularly good at showing the advantages of this aid, and that different configurations might show different results. That sentence still does not appear in the marketing. None of this makes the device poor. It means the evidence is company-funded and company-authored, and that at $6,800 you are entitled to notice.

Update, September 25, 2026: this has changed since the page was last corrected. HearAdvisor, whose standardised bench measurements are the nearest thing this field has to a neutral referee, published its results on Fortell on September 15, 2026 (https://www.hearadvisor.com/products/fortell-ai-hearing-aids; see also Fortell's own summary at https://www.fortell.com/blog/fortell-hearadvisor-press-release). Testing on an acoustic manikin gave Fortell an overall SoundScore of 6.0, a speech-in-noise score of 6.4, which HearAdvisor describes as the highest it has recorded, a first-fit speech-in-quiet score of 5.4, and a perfect result on feedback management. That is a genuine independent measurement, and it closes the gap we described in an earlier version of this page. It still reflects laboratory conditions on a manikin at first-fit rather than an individually tuned real-world fitting. The trade-press gap has narrowed but not closed: there is still nothing in Hearing Review, nothing in The Hearing Journal, no review from Doctor Cliff, none from HearingTracker's Matthew Allsop. Most of the wider professional discussion is still happening in a single forum thread.

The exclusions. Three, and for some readers they settle it before performance enters into it. There is no telecoil, so if you use loop systems the device cannot receive them, and no amount of processing substitutes for that. The app is iPhone-only with no Android version, which is a hard exclusion rather than an inconvenience. And Front Voices, the noise-elimination mode the marketing is built around, has to be switched on by hand rather than engaging when a room turns difficult. Then the one that surprises people most: Fortell is out of network with every insurer. It will hand you paperwork to submit yourself and nothing further. If you hold hearing aid benefits through an employer plan, a union, or a Medicare Advantage rider, that is real money the model gives you no way to use.

What is good about it. We would rather say this plainly than bury it. Building custom silicon to get latency low enough for source separation is not a thing companies do for marketing reasons. The advisory board, which includes Stanford and NYU Langone audiology, is not decorative. The price is honest in a way much of this industry is not: $6,800 covering five years of care, a three-year warranty and a hundred-day trial is a clear bundled number, and plenty of practices charge more for less while being vaguer about it. And the most useful single data point we found is not Fortell's. An independent audiologist described fitting a patient with these devices, then running real-ear measurement and verifying directionality, and reported the result was accurate and the patient delighted. One case proves nothing statistically. But it is exactly the check we would run, done by someone with no stake in the answer, and it came out well.

What we can and cannot do. We cannot sell you a Fortell device and we cannot service one; the company does not supply practices like ours, by design. Two things we can do cost you nothing. If you are holding a quote from anywhere, Fortell included, we will give you a second opinion on it, which means a conversation about whether the device suits your audiogram and the rooms you actually struggle in, not a pitch for something else. And if you already wear aids fitted elsewhere, we will verify them with real-ear measurement: a probe microphone in the canal measuring what the device delivers against your prescription across pitches. First-fit settings routinely land well under target, and an unverified premium aid can easily be beaten by a carefully verified mid-range one. That check is worth having whoever fitted you. We fit Phonak, Oticon, Signia, Starkey, ReSound and Widex at Harrison and Somers, which is the real difference between an independent practice and a single-product clinic: no manufacturer's marketing can wrong-foot us, because none of them pays our rent.

Corrections. We went back over this page against the primary sources and found four statements that were wrong. We have rewritten the relevant passages rather than quietly adjusting them. We had said we could not find the May 2026 preference study indexed anywhere outside Fortell's site; it is published in the International Journal of Audiology at https://doi.org/10.1080/14992027.2026.2663345. We had said neither study states a sample size; both do, in their abstracts, at thirty and twenty. We had said the headline improvement figure moved from 9.2 dB to 17 dB; it did not, and the two are different measurements reported in the same paper. We had said the ten-times claim was contradicted by Fortell's own 80 per cent against 31 per cent figures; those are the 0 dB condition, while the claim concerns minus 6 dB, where the modelled odds ratio is 18.9 and the raw averages are 50 per cent against 7 per cent. We have described the original wording here rather than deleting it, because a page that holds a company to an evidentiary standard should show its own corrections in the open. What had not changed as of that correction: no independent laboratory had published measurements of this device, both studies are funded and authored by the company, and neither names the devices it was compared against. Corrected 8 September 2026. Update, 25 September 2026: the independent-measurement gap noted above has since closed. HearAdvisor has published standardised laboratory test results for Fortell, described above under 'Following the evidence.' This is new information rather than a correction of an earlier error; both studies behind Fortell's own speech-understanding and preference claims remain company-funded and company-authored, and neither names the devices it was compared against. Updated 25 September 2026.