Every AI Car-Sound Diagnosis Tool Compared: Prices, Terms, and What Each Cannot Do

Before a single comparison: this page was written by PulsCar, one of the tools listed on it. That is a conflict of interest, not a disclaimer to skim past, and the only useful response is to make everything here checkable by you in ten minutes with the vendors' own sites open. Second: we did not run a same-recording head-to-head test of these tools. No blind trial, no scored bake-off, no shared clip fed to all eight, and none from anyone else that we could find. So this page contains no accuracy verdict and no accuracy ranking, and it will not have one until somebody runs a test worth citing.

What survives those subtractions is a map of the whole category on the axes you can verify: what each tool costs and how the pricing works, what you upload and what comes back, who is behind it and how to reach them, what refund and privacy terms it publishes, and — the section most drivers should read first — what each tool states, in its own words, that it cannot do.

As of September 2026 at least eight live tools take a recording of a car noise and hand back an AI opinion, run by everyone from a single unnamed individual in Québec to a Pennsylvania company selling into auto auctions. Several are genuinely free. Several publish accuracy percentages nobody outside the vendor has checked. One publishes a price in its Terms that its own homepage contradicts. Three publish no price at all.

I built PulsCar after spending roughly $6,000 over six months in 2025, in Phuket, chasing a noise several shops each diagnosed differently, starting with a $380 "full diagnostic" — and the sound never went away (the story is on the About page). So I have a direct stake in one column below, and the way I have tried to earn your attention is to give every other column the same detail and the same honesty about limits — including a gap in PulsCar's own published terms that two competitors do not have.

One rule overrides everything below: a sound tool answers "what is that noise," not "why is that light on."

How the marks work: two levels, no third

🟢 Verifiable means the fact is printed on the vendor's own page, app-store listing, Terms or Privacy Policy, and you can open that page today and see it: prices, feature lists, policy sentences, company names, retention statements, disclaimers. Every 🟢 here was re-read on 5 September 2026.

⚪ Vendor-stated means the vendor asserts something about its own performance or scale that nobody outside the vendor has checked: accuracy percentages, diagnosis counts, star averages, training-set sizes. A ⚪ is a label, not an accusation — the number may well be true. It means we cannot show you the work.

There is no third level, because a third level would have to be filled with our judgment dressed up as measurement. PulsCar's own unverifiable claims carry ⚪ exactly like everyone else's, and PulsCar is assigned no accuracy percentage, no user count and no star rating anywhere on this page.

What is an AI car-sound diagnosis tool, and what actually separates one from another? It is a service that takes a recording of a car making a noise, runs it through a model, and returns a likely mechanical cause with some notion of urgency. The category splits on four verifiable lines rather than on quality. First, price model: free with no account, free with a paid tier, pay-per-use credits, subscription, or no consumer price at all. Second, delivery: an on-page result in seconds versus a report by email. Third, breadth: some cover brakes, suspension, drivetrain and exhaust, while others are framed around the engine. Fourth, published terms — refunds, retention, and whether any human being is named. Every tool below is placed on those four lines and on nothing else. PulsCar sits in the free, emailed, engine-framed corner of that map, and this page marks its limits as plainly as everyone else's.

The roster: every live tool in the category, September 2026

MyAutoSound — browser tool. Free unlimited text diagnoses; audio costs one credit, packs from $1.99. Run by an unnamed individual in Québec.

Carithm — free browser tool, no signup, on a wider site whose main product is a repair-quote checker. Screens eleven issue areas, returns a ranked list with confidence.

FixMyCarSound — free browser tool, no account, names one of 19 conditions. Terms name NovitasVentures, LLC and Colorado law. Not to be confused with FixMyCar / whocanfixmycar.com, a UK garage-booking marketplace and an entirely different company that does no sound analysis at all.

EngineScan AI — iOS and Android app by Fuat Yahsi of Ankara, Türkiye. Basic result behind a rewarded ad; iOS sells a "Premium Report" for $2.99.

AutoSonix — AutoSonix, LLC of Whitehall, Pennsylvania. Its real product, PTX, is dealer- and auction-facing and combines acoustics with OBD-II data. B2B-first; a consumer app exists but is effectively unreviewed.

EngineEar AI (on Google Play as "AI Car Engine Sound Analyzer") — Android app by alceit of Böblingen, Germany. Analyses need a subscription whose price the listing does not publish.

AI Mechanic: Engine Scan — Android app by Smart Auto Lab of Wrocław, Poland. The largest install base here, and the only panel declaring data is not encrypted.

PulsCar — our own tool. A free 30-second engine sound check, no card, emailed report in ten minutes. The code half needs a scanner; PulsCar's app works with the OBDLink MX+, so we have a commercial interest in it, as the PulsCar section below repeats.

Full comparison: every checkable fact, side by side

Read the table for the pattern rather than for a winner. The tools publishing most about their own limits are not the ones publishing the biggest accuracy numbers. FixMyCarSound names the conditions it is worst at and prints a top-3 figure beside its top-1; Carithm publishes no percentage and instead publishes a behavior, reporting uncertainty rather than guessing. And three of the eight do not publish a price, an operator address, or both — the most avoidable gap in the category.

What actually determines which tool you should use

Whether you can record the noise at all. The master variable, and the one nobody discusses. MyAutoSound's free tier works from a written description — "metallic clicking, front left, only when turning right" — so it returns something when the noise refuses to perform for a phone. Every other tool needs a clip, and bad technique is the largest single cause of a useless result; the recording guide fixes more results than switching tools does.

Where the noise lives on the car. MyAutoSound labels six categories including brakes, suspension, drivetrain, exhaust and HVAC; FixMyCarSound's 19 conditions span the same ground; Carithm's eleven areas include brakes, bearings and suspension. PulsCar's check is framed around the engine — "film your engine for 30 seconds" — so a brake squeal has a labeled home elsewhere and not here. That is a scope fact, not a quality one, and it should decide your first click.

How fast you need it. MyAutoSound and Carithm return on the page in seconds; FixMyCarSound quotes about 60 seconds end to end; PulsCar emails within about ten minutes. In a parking lot deciding whether to drive home, seconds matter; at a kitchen table deciding whether to keep the car, they do not.

Whether the price is knowable before you commit. This separates the category more sharply than accuracy does. MyAutoSound publishes five pack prices; EngineScan AI publishes $2.99 on iOS; FixMyCarSound's homepage says free while its Terms carry a $9.99 pass; EngineEar AI and AI Mechanic publish nothing. A tool you cannot price before installing is a tool you cannot compare.

What the vendor publishes about your recording. MyAutoSound says diagnosis audio is deleted immediately after processing and names its processors; FixMyCarSound says audio lives 90 days on encrypted S3 and records 12 months; two Android listings claim immediate deletion while their own data-safety panels say otherwise; Carithm and PulsCar publish nothing on the point.

How much an unverified percentage moves you. A reader reassured by "93%" will gravitate to MyAutoSound; a reader unsettled by an unaudited number will prefer the tools publishing none, or the one naming its weak classes. Both instincts are defensible, and neither tells you which tool is more often right.

The price ladder: every published number, read 5 September 2026

Every row was read from the vendor's own page that day.

The ladder's shape is the honest headline. Everything a driver with one noise actually needs sits in the top eight rows, between $0 and $2.99. Against what a shop charges just to look at the car, each is a rounding error. The bottom two rows are different: a 72-hour pass whose own homepage says the tool is free, and three prices that are not published at all. Note also what the ladder does not contain — a row where paying more buys measurably better accuracy, because no such measurement exists.

How should you actually spend money here, given that nobody has tested accuracy? Spend nothing first. Four tools return a real answer at zero cost with no card, so the rational first move on any noise is to run it through two of them and see whether they agree — agreement between independent systems is the closest thing to a second opinion available without a lift and a mechanic. If they disagree, or the noise worries you enough to want a paid pass at it, MyAutoSound's smallest credit pack is $1.99 and EngineScan AI's Premium Report is $2.99, roughly one percent of a typical shop diagnostic fee. Reserve real money for the physical inspection, not for the app. PulsCar's own check stays at zero for the same reason: at this stage of the category, charging for a first opinion nobody can audit is the wrong ask.

What each tool says it cannot do, in its own words

The most useful thing a diagnostic tool publishes is its own boundary. Read this before any accuracy figure.

On OBD-II codes — the universal limit. Neither a sound tool nor any app can read a stored trouble code from a recording, because the code has no acoustic signature. FixMyCarSound is the bluntest, under a heading saying exactly that: "It cannot feel for play in a wheel bearing, smell a coolant leak, check compression, or read trouble codes. It cannot tell you whether the sound is coming from the left front versus the right rear, and it cannot inspect brake pad thickness, rotor condition, or CV boot integrity." MyAutoSound routes around it by recommending hardware — its "Recommended gear" section links an OBD2 scanner so you can "read and clear basic engine codes at home and match them with what MyAutoSound suggests." PulsCar draws the line in a sentence — "An OBD scanner reads error codes. PulsCar listens" — and marks OBD codes outside the free check's detection scope. AI Mechanic calls itself "a powerful alternative to OBD2 scanners for physical engine wear," with the qualifier doing all the work. AutoSonix is the exception that confirms the rule: PTX "cross-checks what it hears against the car's own onboard diagnostics" precisely because acoustics alone are incomplete. A warning light on a car that sounds normal is the wrong category for every tool here — the sound-versus-scanner guide covers what to use instead.

On being a diagnosis at all. MyAutoSound: "general guidance only — not a certified diagnosis," and in its Terms, "A paid result is still an informational output only and is not a guaranteed diagnosis." Carithm's footer: a "decision-support tool. Not a substitute for physical inspection." EngineScan AI: "It does not replace a professional inspection, service diagnosis, or safety decision." EngineEar AI: "provides informational assessments only." AI Mechanic: "It is not a replacement for a certified mechanic." AutoSonix: "A PTX score isn't a diagnosis." PulsCar: "PulsCar isn't a replacement for major repairs." Seven vendors, one sentence, seven voices.

On what a bad recording does to the answer. Here the tools genuinely differ. Carithm publishes a behavior rather than a percentage: "If recording quality is insufficient, the system says so — it never invents a diagnosis," and, asked whether the AI always makes a prediction, "No. When evidence is insufficient, the platform reports uncertainty rather than generating unsupported conclusions." FixMyCarSound warns that "Background noise and distance significantly affect accuracy." MyAutoSound's Terms concede that "Audio analysis results depend on the quality, length, and clarity of the submitted recording." PulsCar returns "inconclusive" instead of guessing. Each is a promise about what the tool does when it does not know, and that promise is worth more than a headline number.

On the specific things sound cannot separate. FixMyCarSound is the only tool publishing its weak classes by name, the most useful disclosure in the category: vacuum leaks, accessory belt noise and transmission pump whine "hit a physics-based ceiling and are best supplemented with OBD-II data." It also states that a result pointing to rod knock, timing chain failure or brake wear means "do not keep driving to see if it gets worse." MyAutoSound publishes its own urgency list — grinding, strong vibration, brake, steering and power-loss symptoms — and says to stop driving if the noise "feels unsafe or worsens quickly." If your symptom is on one of those lists, the app's job is to send you to a lift.

The eight tools in depth

MyAutoSound — the cheapest audio analysis, and the clearest terms

MyAutoSound is the most fully built consumer tool here and the only one with a genuinely free text mode: type what you hear and get a main cause, an urgency level, a "Can I drive?" verdict and basic safety checks, unlimited, no card. Audio analysis costs one credit and adds acoustic analysis, a mechanic-ready PDF, a cost estimate, differential causes and red flags, and a 30-day share link, across six categories and "50+ noise types."

The transparency picture is two-sided. On terms it is the best-documented tool in the comparison: Terms and Privacy Policy both last updated 19 June 2026, an explicit no-refund clause, named processors including OpenAI, Firebase and Stripe, a stated deletion policy for diagnosis audio, and Québec law with Montréal courts. On self-reporting it is the most aggressive: "93% Accuracy (audio mode)," "5 000+ Diagnoses completed," "4.8 ★ Average rating," "50 000+ engine recordings," and named five-star testimonials carrying, in smaller text on the same page, the line "Testimonials are illustrative examples." All of that is ⚪. For the two-tool detail line by line, see MyAutoSound vs PulsCar.

Carithm — free, no signup, and unusually honest about uncertainty

Carithm is the lowest-friction entry point in the category: open the page on a phone, upload or record, get a ranked list of likely fault areas with a confidence score in a few seconds, no account. It screens eleven areas, from engine knocking and valve-train ticking through belt squeal, wheel-bearing humming, brake grinding, suspension clunks and transmission whine to AC compressor noise.

What sets it apart is a published refusal rather than a published percentage. The page says the system "evaluates recording quality, measures uncertainty, and explains when additional inspection is recommended," that it "never invents a diagnosis when the recording is unclear," and, asked whether it always makes a prediction, answers "No." In a category whose loudest signal is usually a two-digit accuracy claim, publishing the conditions under which a tool declines is the more falsifiable move. The gap is on the other side: as of 5 September 2026 the page carries no operator name, no contact route, no privacy policy and no statement about what happens to your recording.

FixMyCarSound — the most detailed limits page, and a price contradiction

FixMyCarSound records up to 15 seconds in the browser, asks a well-designed set of context questions — when it happens, whether pitch tracks engine revs or road speed, whether it worsens under load — and returns one of 19 named conditions with a severity band, a repair-cost estimate, a top-3 alternatives list and a PDF. Its FAQ explains that those answers act as "hard gates to eliminate impossible diagnoses," which is a real diagnostic idea rather than marketing.

It publishes the best limits page in the category, quoted in full above, and the most falsifiable accuracy shape: "76% top-1 · 88% top-3," with twelve conditions named as strongest and three as hitting a physics ceiling. All of it remains ⚪ — there is no methodology page. Two cautions. The price contradiction: the live tool and FAQ say free with no card, while the Terms carry a $9.99 per 72-hour pass and an all-caps no-refund clause; we report both and take no view on which governs. And the Terms name NovitasVentures, LLC and Colorado law, then leave the business address as a bracketed placeholder. The brand collision is worth naming too: searching "fix my car" returns FixMyCar / whocanfixmycar.com, a UK marketplace that books you into one of its garages — different company, different country, no sound analysis.

EngineScan AI — the cheapest paid report, on both stores

EngineScan AI is the only tool here shipping on both stores under one developer, Fuat Yahsi, with a published street address in Ankara, Türkiye and a support email — more identity disclosure than three of the browser tools manage. The flow is short: start the engine, hold the phone near the bay, record about five seconds, and get a three-level result with a confidence percentage, an urgency band from Low to Urgent, recommended actions and a shop locator, across nine listed fault types.

Two verifiable frictions. The Play listing states under PRIVACY that "We do not collect personal data" and "Audio recordings are not stored on the server," while the same page's data-safety panel says the app may collect Location, Financial info and four other types and may share Location, App activity and two others. Both sit on the vendor's own pages, so both are 🟢, and the tension between them is the fact. Second, the developer website linked from the App Store loaded for us as a near-empty page.

EngineEar AI — a maintenance app that also listens

Listed on Google Play as AI Car Engine Sound Analyzer and branded EngineEar AI, this is the most feature-dense entry in the category and the one with the clearest developer identity: alceit, a named developer in Böblingen, Germany, with a published phone number and email. The recording protocol is the most demanding here — a guided mode of about 50 seconds, 30 at idle plus 20 on revs — and the output adds something nobody else offers: a wear evaluation comparing your engine's signature against what a car of that make, year and mileage should sound like.

The pricing model is the problem for a comparison page. The listing says garage features are free and that "A subscription is only required for new AI analyses," then defers every number to in-app screens. That is legitimate and common on Play, and it makes the tool impossible to place on a price ladder without installing it — which is why it sits in the "not published" row rather than getting an invented figure. The same duality shows on data: the description promises audio "deleted immediately after analysis," while Play's data-safety panel reports that audio may be shared with third parties and that data cannot be deleted.

AI Mechanic: Engine Scan — the biggest install base, the weakest data panel

AI Mechanic: Engine Scan, from Smart Auto Lab of Wrocław, Poland, has the largest published install base of the app-store entries at 10,000-plus and the shortest listing. It pitches itself as "Just like Shazam for music," offers an instant acoustic scan with a repair estimate, a seven-day health tracker, and one usefully specific promise — distinguishing rod knock from lifter tick, the highest-stakes call in engine noise. The disclaimer is present and correct: "It is not a replacement for a certified mechanic."

Everything else is unpublished: no price, no accuracy claim, no methodology, and a data-safety panel that simultaneously reports "No data collected" and that Audio is shared with third parties, with data "isn't encrypted" and "can't be deleted." Those are the store's own fields, filled in by the developer, which is why this tool is listed as a member of the category rather than recommended for a job.

AutoSonix — the serious one, and it is not selling to you

AutoSonix is the only company here with a full corporate footprint: AutoSonix, LLC, a street address in Whitehall, Pennsylvania, a phone number and a business email, all published. Its product, PTX, records a vehicle at intake, cross-checks the acoustic signal against the car's own OBD readings, and returns a single mechanical score from 0 to 100. The framing a consumer should borrow is the section it calls the DTC gap: "No warning light doesn't mean no problem... OBD only reports what a sensor has already caught. Most mechanical wear announces itself in sound first, often long before a code ever sets." That is the argument for this whole category, written by the company least interested in selling to individuals.

Say the rest plainly: this one is B2B-first. There is no consumer pricing page, the audience list is auctions, dealerships, banks and warranty providers, and the site says "A PTX score isn't a diagnosis." A consumer AutoSonix app does exist from AutoSonix LLC, listed free, and the App Store reported it had not received enough ratings to display an overview. Press coverage is indexed under Automotive News; that page returned an access error to us on 5 September 2026, so we do not cite its contents. If you are a driver with a noise, this is a company to know about and not a tool to use today.

PulsCar — free, emailed, engine-framed, and willing to say inconclusive

Full disclosure: this is my product, so check every sentence against the site. PulsCar's check asks for a 30-second video of the engine running and emails a report within about ten minutes: an Engine Health Score from 0 to 100, a system-by-system breakdown, an independent AI second opinion reviewing the first pass, a six-month prediction and repair-cost ranges. No card, no tier, no subscription.

The published behavior I would point a skeptical reader at is what the site calls the Honesty Guarantee: "If your clip can't support a confident verdict, your report honestly says 'inconclusive'." That has a cost — some people get a report declining to name a part — and it is why we publish no accuracy percentage. The limits belong in the same paragraph. There is no text mode, so a noise you cannot capture gets you nothing. The check is engine-framed, so a brake or suspension noise has a labeled home at MyAutoSound, FixMyCarSound or Carithm and not here, and electric vehicles are out of scope. And as of 5 September 2026 the site publishes no retention statement for the recordings you send, which MyAutoSound and FixMyCarSound both do — a gap on our side, listed because a table that hid it would be worthless.

One more disclosure belongs in this section. The PulsCar app pairs with a scanner to read trouble codes and live data — the code half that no tool on this page can reach from a recording — and PulsCar's app works with the OBDLink MX+, so we have a commercial interest in it. Nothing on this page needs it, and the free check stays free.

Which tool should you open first, given everything above? Start with the free ones, and start with two of them, because the only quality signal available today is whether independent systems agree. If the noise is at a wheel, a brake or a strut, Carithm and FixMyCarSound both have labeled categories for it and both cost nothing. If you cannot record the noise at all, MyAutoSound's free text mode is the only path here that works from a written description. If the noise is from the engine bay and you want a scored report you can sit with rather than a verdict in seconds, PulsCar's check is free and arrives by email in about ten minutes. Escalate to a paid result only when the free ones disagree, or when a shop has quoted four figures. PulsCar's check is one of the free ones, and it stays free whether or not you ever buy anything from us.

What we re-read on 18 September 2026, tool by tool

A category map goes stale faster than anything else on a blog, so every tool here was opened again on 18 September 2026, thirteen days after the readings above. This is the diff, including the tools that did not move.

MyAutoSound — prices identical, one policy newer, one page down. The credit packs are unchanged at 2 for $1.99 through 15 for $13.99, still "No subscription ever," still "Credits never expire." Its Privacy Policy now shows a last-updated date of 12 September 2026 and names more processors than before: OpenAI, Firebase and Google Cloud, Stripe, Render for backend hosting, Google Maps, Places and Analytics, and OpenStreetMap with Nominatim and Overpass. The Terms page — and two obvious aliases — returned a maintenance notice on the day we looked, so the refund quote on this page stays dated to our earlier reading rather than presented as checked today. Two new things on the homepage: an analyzer widget labeled "Next-gen sound engine — Beta · in training," with the note "Training a new, more accurate engine — try the mic now, it won't identify the sound yet," and a disclosure under the three five-star testimonials reading "Examples above are illustrative, not verified customer quotes."

Carithm — unchanged, and still the thinnest legal footprint. Free, "No Signup," "No App Needed," analysis in about five seconds, eleven issue areas, a ranked list with a confidence score, and the same two sentences that make it interesting: "it never invents a diagnosis" when the recording is poor, and "When evidence is insufficient, the platform reports uncertainty rather than generating unsupported conclusions." There is still no operator name, no contact address, no privacy policy and no terms — only a "© 2026 Carithm" footer. For a tool you send audio to, that is the gap worth knowing about, and it is the same gap as two weeks ago.

FixMyCarSound — the price contradiction survives, and a bigger one appeared. The homepage still says free, no account, 19 conditions, up to 15 seconds of audio, and still publishes the category's most falsifiable accuracy line: "Overall accuracy: 76% top-1 · 88% top-3 · Near-100% on high-confidence classes." The Terms, last updated 17 February 2026, still carry "Current pricing: $9.99 per 72-hour diagnostic pass" and "ALL PAYMENTS ARE FINAL AND NON-REFUNDABLE under any circumstances," with Colorado law, binding individual arbitration with a 30-day opt-out, and liability capped at what you paid in the preceding 30 days "or $50.00, whichever is less." The retention story is in the next section, because it is worth its own.

EngineScan AI — actively maintained, barely rated. The iOS listing shows version 1.1.4 updated within the past week, a single in-app purchase priced at $2.99 ("Premium Report"), developer Fuat Yahsi, 41.4 MB, Utilities, ages 4+, five languages, and a rating of 5.0 from two ratings — a number small enough that it should not move anyone's decision. Its privacy summary lists location, identifiers and usage data collected for tracking and advertising, with "Audio Data" collected but not linked to identity. Frequent updates are a good sign about upkeep; two ratings say nothing about quality in either direction.

EngineEar AI — 5,000-plus installs, and the same panel contradiction. Updated 10 August 2026. The listing states "ALL GARAGE FEATURES WORK WITHOUT A SUBSCRIPTION. A subscription is only required for new AI analyses," with weekly and monthly plans plus a one-time Boostpack, and "Current prices and usage limits are shown in-app before you subscribe" — so the price is still unknowable before installing. Recording is about 50 seconds in guided mode (30 seconds at idle plus 20 seconds of revs) or up to 90 seconds free. The description promises "Audio deleted immediately after analysis, GDPR compliant" and "analysed, deleted from our servers within seconds"; the Play data-safety panel on the same page says the app may share "Audio, App activity and 2 others," that "Data is encrypted in transit," and that "Data can't be deleted." Both statements are the vendor's own, and they do not agree.

AI Mechanic: Engine Scan — the largest install base, the strongest claims, the weakest panel. Still 10,000-plus downloads, updated 23 July 2026. Its data-safety panel says the app may share Audio with third parties, declares "No data collected," and then states "Data isn't encrypted" and "Data can't be deleted" — the only listing in this comparison declaring no encryption. Its description calls the app "a powerful alternative to OBD2 scanners for physical engine wear," compares itself to Shazam, and claims it can "Accurately distinguish between minor top-end noise and fatal bottom-end engine damage." One thing did improve since our earlier reading: the "About the developer" block now publishes a named individual, a street address in Wrocław, a phone number and a support email — Google Play requires that disclosure, and this listing now carries it, which several tools on the open web still do not.

AutoSonix — unchanged, and still not selling to you. The site publishes a full corporate footprint: AutoSonix, a street address at 3848 Lehigh St, Suite C, Whitehall, Pennsylvania 18052, a phone number and [email protected]. PTX still "listens to a vehicle, cross-checks what it hears against the car's own onboard diagnostics and other contextual data, and delivers one objective mechanical score in minutes, at intake," and the audience list is still auctions, fleets, dealerships, CPO operations, banks, lease companies and warranty providers. The sentence we keep quoting is still there: "A PTX score isn't a diagnosis. It's the starting point every decision has been missing." No consumer price anywhere.

PulsCar — unchanged, including the gap we keep printing. The free check is still free with no card, still a 30-second video of the engine running, still emailed, and the homepage still carries no retention statement for the recordings you send. That gap is ours.

Three pages, three retention stories

FixMyCarSound deserves a section rather than a row, because it is the clearest available lesson in how to read any of these tools — including ours. Three of its own pages describe what happens to your recording, and they do not say the same thing.

The homepage says: "Your recording is only used for this diagnosis and is not stored permanently."

The Privacy Policy says: "Audio recordings are retained for 90 days, then permanently deleted from AWS S3," with diagnosis records kept 12 months, storage in AWS S3 with server-side encryption, and a right to request deletion by email with a response within 30 days.

The Terms say: submitted audio and associated data "may be Stored indefinitely," and that you grant a "perpetual, worldwide, royalty-free license to use, store, analyze" it.

All three were read on 18 September 2026. We are not calling this deceptive: the likeliest explanation is that a homepage line, a privacy policy and a boilerplate license clause were written at different times by different hands, which is ordinary for a small product. A reader trying to answer "what happens to my recording" gets three answers, and the one that governs in a dispute is the one in the Terms, not the friendly one on the homepage. That ranking — Terms over Privacy Policy over marketing copy — is how to read every vendor in this category.

For contrast, the two ends of the same spectrum: MyAutoSound's Privacy Policy states that diagnosis audio is "deleted immediately after processing" and names every processor that touches it, which is the strongest disclosure here; Carithm publishes no policy at all, and PulsCar's homepage publishes no retention sentence, which is why both carry that gap in the table above rather than a reassurance.

The practical rule: if a recording of your car matters to you, read the Terms clause on licensing before you upload, not the homepage. And if you would not want a clip kept indefinitely, prefer a tool that states a deletion time and names where the file lives.

The traps: three ways to get a bad outcome from a good tool

Trap one: reading a vendor number as a measurement. The situation: a driver picks the tool with "93%" on the homepage, or the one publishing "76% top-1 · 88% top-3" because it looks scientific, or the one publishing nothing because silence reads as modesty. What is real: none of those signals is evidence. No independent test of any tool here exists that we could find, no vendor publishes a methodology page, and the presence or absence of a percentage tells you about a marketing decision, not a model. The price: overconfidence in either direction, applied to a four-figure repair decision. The defense: treat every output as a first opinion and run a worrying noise through two independent tools. If one result is about to authorize real money, the next step is a lift, not a second app.

Trap two: the price you could not find before you needed it. The situation: a driver installs an app that says "In-app purchases" with no figure, records the noise, gets a teaser result, and only then meets a weekly subscription — or reads a homepage that says free and later finds a $9.99 pass in the Terms. What is real: three of the eight tools here publish no price before installation, and one publishes two different prices on its own site. That is a comparison problem rather than necessarily a scam, but the cheapest-looking tool in any list is often just the one that has not told you yet. The price: a subscription renewing for a tool you used once. The defense: check the store listing for a price range and the vendor's Terms for a pricing clause before installing.

Trap three: the recording that was never going to work. The situation: a noise that only appears at 40 mph over rough pavement gets recorded at idle in a driveway, from inside the cabin, with the radio on. What is real: the sound the model needs is not in the file, and every vendor says so — FixMyCarSound warns that "Background noise and distance significantly affect accuracy," MyAutoSound's Terms note that results "depend on the quality, length, and clarity of the submitted recording," and PulsCar returns "inconclusive." The price: a credit spent on nothing, or a report about a healthy idle read as an all-clear on a failing bearing. The defense: reproduce the exact condition, get the microphone near the source, and follow the recording guide before spending anything.

Three composite situations, decoded

These are composite scenarios, not individual customers — the shapes of request the category is built around, written in the subjunctive on purpose.

Scenario 1: the parking-lot squeal with five minutes to decide. Suppose a driver hears a squeal on braking, is standing beside the car, and needs to know whether to drive home. The free browser tools would fit: they return on the page in seconds, both FixMyCarSound and Carithm carry brake categories, and MyAutoSound would add a "Can I drive?" verdict for nothing in text mode. PulsCar's emailed engine check would be slower and category-mismatched here, and this page should say so rather than route the reader to its own product. Lesson: immediacy and a labeled category decide this one, not accuracy claims.

Scenario 2: the keep-or-sell knock on a high-mileage car. Suppose an owner past 140,000 miles hears a faint knock that rises with engine speed and is deciding whether to spend money or move the car on. Output shape matters more than response time: a scored report with a system-by-system breakdown and a six-month outlook is built to inform a decision, while a one-line verdict is built for a shop visit. Neither settles it alone: a knock that rises with engine speed is the case where a shop's compression or leak-down test, or a stethoscope on the block, is worth its fee before money changes hands either way. Lesson: match the output shape to the decision you are making, not to the speed you can get it.

Scenario 3: the used car on a test drive. Suppose a buyer twenty minutes into a test drive hears something the seller has not mentioned. A browser tool would give an on-the-spot read to raise at the curb, and a wear-comparison feature of the kind EngineEar AI describes is aimed at exactly this moment. What none of them can do is tell you whether the pad thickness or the CV boot is intact — FixMyCarSound says so in as many words. Lesson: a sound tool is a negotiating input and an early warning, never a substitute for hands on the car.

Which tool is yours? Answer four questions

Question 1: Can you record the noise clearly at all? → No → MyAutoSound's free text mode, the only path here that works from words. Yes → continue.

Question 2: Where on the car is the noise? → Brake, wheel, suspension, exhaust, HVAC → Carithm, FixMyCarSound or MyAutoSound, which carry labeled categories. Engine bay → any of them, plus PulsCar's check.

Question 3: Do you need an answer in the next sixty seconds? → Yes → a browser tool with an on-page result. No → an emailed report buys a second-opinion pass.

Question 4: Would you install an app whose price is not published? → No → stay in the browser, where four tools publish "free" or a full price table.

Four sound-diagnosis tool playbooks for where you are right now

"There is a noise and I want an answer in five minutes." Record close to the source with no radio, reproducing the condition where the noise is loudest, then run the clip through two free browser tools. If they agree, you have something worth repeating to a mechanic; if they disagree, the next step is a physical inspection, not a third app.

"There is a noise and I want a report I can sit with." Film 30 seconds of the engine running for PulsCar's free check. If it comes back "inconclusive," treat that as instruction to re-record rather than as good news, then cross-check the named component against the noise index.

"A warning light is on and the car sounds completely normal." None of these tools. A stored code has no sound; every vendor here says so, and two recommend a scanner explicitly. Read the code with an inexpensive reader or a free parts-store scan — the diagnostic-app comparison picks the reader.

"A shop quoted four figures for a noise I recorded." Run the recording through two independent tools for under three dollars total and see whether either names the component on the quote, then check the number against what noises typically cost to fix. Two reads naming a different, cheaper part are not a verdict, but they are a reason to ask the shop to show its diagnostic steps before you authorize work.

A code reader is the other half of this category, and its economics work differently: what FIXD reads and what it costs sets out the four different yearly prices FIXD publishes on its own pages, and counts how many owner complaints a code reader would never have seen.

After you pick: four habits that beat switching tools

Record once, use everywhere — one good clip works in every tool here, a bad one fails in all of them. Read the boundary before the answer: the urgency band decides whether you drive tonight. Never let a clean sound report clear a warning light. And spend on disagreement, not confidence: two tools naming different parts is the finding, and it belongs at a lift.

What this category is actually aimed at: 55,169 noise complaints, by system and by word

Every tool above publishes a coverage claim — "50+ noise types," "19 automotive conditions," "11 issue areas" — and none of them publishes what the field looks like. So we measured it, from the public NHTSA complaint file: 777,040 owner complaints received between 1 January 2015 and 8 September 2026, of which 55,169 describe a noise affirmatively.

Which systems do those complaints belong to? NHTSA assigns its own component labels, and a complaint can carry more than one:

Read that as a map of where the category's demand sits, not as a ranking of what makes noise. An engine-framed tool — ours included — is aimed at the largest single label and roughly a third of the file. A tool carrying brake, suspension and steering categories is aimed at a comparable third that an engine-only check will not name. Neither framing is better; they are aimed at different complaints, and that is the most useful thing this page can tell you about scope.

Which words do owners actually use? Counted across all 777,040 complaints, with each word required to sit within 100 characters of a sound word so that "a message popped up" does not count as a noise:

Knocking leads at 5,777 complaints, then grinding 4,956, clicking 4,807, rattling 3,714, clunking 3,128, ticking 2,281 and whining 1,766. Popping is nominally larger at 5,963, and it is the one number in the table we would not lean on: before the sound filter it stood at 13,639, mostly warning messages popping up on a screen, and much of what survives describes a one-off event — a tire blowout, a rear window shattering — rather than a repeating noise a classifier could learn. At the other end, the words that look most vivid are the rarest: growling 164, groaning 123, chirping 201. Download both tables; the method note gives every word form, both filters, the sample reading and what the filters removed.

Two conclusions a reader can use. First, a coverage claim of "19 conditions" is not obviously thinner than "50+ noise types," because the vocabulary owners actually use is concentrated: seven words account for most of what people write down. Second, and more important for choosing a tool, the noise word you would type is a weak signal about the system at fault — clicking appears in steering, driveline and starter complaints alike — which is exactly why every vendor here asks for the condition as well as the sound, and why the recording guide spends more space on when the noise happens than on which word describes it.

As always with this data: complaints are unverified, self-selected owner reports, they are not failure rates, they are not normalized to vehicles on the road, and nothing here ranks any make or model. The queries were written case-insensitively because narratives are in capitals through 2020 and mixed case afterward, and a capitals-only query would have silently dropped most of the recent half of the file.

Sources and documents

Every vendor fact came from that vendor's own page or store listing, on the date given. The analyses are ours and downloadable.

• MyAutoSound homepage, pricing and FAQ and Privacy Policy (page states last updated 12 September 2026) — prices, credit packs, accuracy and rating claims, the deletion sentence, the processor list, the beta-engine note and the testimonial disclosure. Read 5 September and re-read 18 September 2026; the Terms returned a maintenance notice on 18 September 2026.
• Carithm — free, no signup, eleven issue areas, the ranked list with confidence, and both uncertainty sentences. Read 5 and 18 September 2026; no terms, privacy policy or operator contact published.
• FixMyCarSound, its Terms (last updated 17 February 2026) and its Privacy Policy — 19 conditions, the top-1 and top-3 accuracy line, the limits paragraph, the $9.99 72-hour pass, the non-refundable clause, Colorado law and arbitration, and the three different retention statements. Read 5 and 18 September 2026.
• EngineScan AI on the App Store — free with ads, the $2.99 Premium Report, developer, version, rating count and privacy summary. Read 18 September 2026.
• AI Car Engine Sound Analyzer (EngineEar AI) on Google Play — install band, update date, subscription wording, recording lengths, the deletion promise and the data-safety panel that disagrees with it. Read 18 September 2026.
• AI Mechanic: Engine Scan on Google Play — install band, update date, the OBD2-alternative and rod-knock claims, the data-safety panel and the About the developer block. Read 18 September 2026.
• AutoSonix — the PTX description, the B2B audience list, the company address, phone and email, and the "isn't a diagnosis" sentence. Read 5 and 18 September 2026.
• NHTSA Office of Defects Investigation complaint flat file — the field dictionary for the public data behind the two tables above: 777,040 complaints, 1 January 2015 to 8 September 2026.
• PulsCar analysis: noise complaints by NHTSA component and owner noise vocabulary, with the method note. CC BY 4.0.
• NHTSA recall lookup by VIN — the free official check for open recalls on your own car, which no sound tool can do for you and which is worth ten minutes before any of them.

Neighbors of this page: MyAutoSound vs PulsCar for the two-tool detail, the full diagnostic-app comparison for the OBD2 half of the market, sound AI versus OBD scanners for why you need both, how to record a car noise for the input that decides every result, strange car noises decoded, car noises ranked by repair cost, and what a diagnostic should cost.

How this page was built: every fact was read from the vendor's own site or app-store listing on 5 September 2026 — myautosound.com including its Terms at /legal and its Privacy Policy, carithm.vercel.app, fixmycarsound.com including its FAQ, Terms and Privacy Policy, both store listings for EngineScan AI, the Google Play listings for AI Car Engine Sound Analyzer and AI Mechanic: Engine Scan, autosonix.ai and the App Store listing for the AutoSonix app, and pulscar.io with its About page. Prices are quoted as published. No same-recording head-to-head test was run by us, and we found none published by anyone else; accordingly this page contains no accuracy ranking. Accuracy, diagnosis-count, rating and training-set figures are vendor statements with no published methodology and no independent verification we could locate; PulsCar publishes no accuracy percentage and is assigned none here, along with no user count and no star rating. A press article about AutoSonix is indexed under Automotive News; that page returned an access error to us and its contents are not cited. Scenarios are composite.

Prices verified September 2026 — every published price in the ladder above was re-read from the vendor's own page or store listing on 18 September 2026 and none had changed; the three tools that publish no price still publish none, and no repair-cost estimator is quoted anywhere on this page. Technical documents verified September 2026 — the vendor Terms and Privacy Policies, both Google Play data-safety panels, the App Store privacy summary and the NHTSA complaint flat file behind the two tables were all opened on 18 September 2026; the one exception is MyAutoSound's Terms page, which was serving a maintenance notice that day, so its refund quote stays dated 5 September 2026. Reviewed quarterly.

Used two of these on the same noise, or found a tool we missed? Email [email protected] with what each one said and what the shop actually found — a real same-recording result from a reader is worth more than anything we could write, and this page gets updated as the category changes.