Runsense Review (2026): The AI Running Coach That Prescribes Effort, Not Pace

A running coach that lives in your messages

Most running apps want you to open them. Runsense wants to talk to you.

You tell it what you are training for, it builds a real periodized plan to get you there, and then it more or less gets out of the app and into your messages. Every run you do arrives on its own. When your week falls apart, you say so in a sentence and it rebuilds around you.

Runsense is an AI running coach for iPhone. You give it one goal with a date, it writes a periodized plan to get you there, and then it coaches you through the plan by message rather than expecting you to open an app. It reads your runs from Apple Health, Strava, Suunto or Polar, it prescribes sessions in effort rather than pace, and it answers questions about your own training in a conversation. One price, no tiers, and a 14 day trial that takes no card.

That is the pitch. The reason it is worth reading about is that the pitch mostly survives contact with the product, and the part that survives best is not the plan. Plan generation is a solved problem. What is not solved, anywhere in this category, is what happens when you ask a question that a chart cannot answer.

First, make sure you are installing the right app

This is the most practical thing in this review, so it goes near the top.

Three unrelated apps in the App Store are called Runsense. The one this review is about is Runsense: Run Coaching, published by Raphael Crawford-Marks. There is a separate Runsense from Different Pen Limited, subtitled “Run faster and further”. And a third, RunSense, from PragmaTec.

The listing above is the right one. Match on both the full name and the developer, because the one you want does not always come up first. Searching from India put Different Pen’s app above Raphael’s. The ordering shifts by storefront.

Two things on that listing are worth reading before you install anything. It is iPhone only, stated by Apple rather than inferred by us. And it carries three ratings. They average five stars, which means nothing at that volume. Three ratings is the honest measure of how new this is, and it is the number you should weigh against everything good we are about to say.

Three ways to be sure:

• The developer line under the title reads Raphael Crawford-Marks.
• The subtitle reads “From first run to finish line”.
• Or skip the search and use the download link on runsense.ai, which goes straight to the right listing.

If what you opened shows you a body map of aches and pains, or motivational quote cards, you have the wrong one.

Setting it up

Setup is four rows, and it takes about a minute.

You give it one goal with a date. That single field is what everything else hangs off, and the countdown to it sits at the top of the app from then on. You point it at wherever your runs already land. You pick how you want to be coached: by text, by email, or in the app. And that is essentially it.

That channel choice is the whole design in one setting. Pick text and the coaching comes to you the way a message from a person would, which is the point: you are not relying on remembering to open anything.

The heart rate row sat empty for us the whole time, which turned out to matter in an interesting way, and we come back to it.

What connects, and how

Two things worth pulling out of that table.

The Polar limitation is Runsense’s own disclosure, not something we caught them out on. They tell you on their own page that connecting Polar will not bring your older history over. Most companies let you find that out yourself.

And you do not have to buy anything to start. A phone in your pocket is a supported setup, not a degraded one. What a watch adds is the recovery half of the picture rather than the coaching itself.

What a week actually looks like

The plan we were given was a 10 week build to a 5K, and the first thing you see when you open the app is the shape of the current week on a single strip.

Easy, strength, workout, rest. Every day is typed, so you never have to work out what today is supposed to be, and the race countdown sits above it so the week always has a reason attached.

Tap into the day and the session is written out in full. Here is where the product’s best small decision shows up: the sessions are prescribed in effort, not pace.

A Wednesday intervals session read: five minutes of brisk walking to warm up, then five efforts of ninety seconds at an intensity where “you can only speak in short phrases, breathing deeper but still in control”, each one followed by a full two minute recovery walk, then five minutes of easy walking to cool down.

You can run that. You do not need to know your threshold, own a watch, or have any idea what your 5K pace is. You check the instruction against your own breathing, mid-run, and you always know whether you have it right.

And underneath it, this:

You should finish the last one looking like the first. If the last is a struggle, the first was too fast.

That is a coach talking. It is the sort of line you get from someone who has watched a lot of people run intervals too hard, and it is not the sort of line a plan generator produces.

Saru says: The real finding here is that prescribing in effort is not a limitation dressed up as a feature. Pace targets are a proxy for effort, and a bad one, because they ignore heat, hills, sleep and how your week went. Runsense skipped the proxy and prescribed the thing itself. That is why it works on a phone with no sensors, and it is also why it is harder to fake.

Around the running sit the extras, and they are marked optional rather than mandatory: a bodyweight strength and mobility routine with a video, and the Myrtl routine, a twelve move mobility circuit with both a PDF and a video.

That framing matters more than it sounds. Mobility work is the thing that keeps beginners uninjured and it is also the first thing everyone drops, so a plan that hands it to you as homework is a plan people quit in week three. Here it sits beside the run, available, and the copy explicitly says taking full rest instead is completely fine.

Two other details in that screen are worth noticing. A gym session that was never part of the plan appears anyway, pulled in from Apple Health and labelled “Also logged, not on plan”. And where the day offers an easy run, it offers alternatives in the same breath: a run, or a slow uphill treadmill walk-jog, which is a small kindness to anyone whose weather or knees are not cooperating.

How much of this plan is actually yours

Tap the mobility routine sitting on your plan and what opens is not something Runsense wrote. It is Nike’s published Myrtl routine, served in the app with its own branding still on it, credited rather than laundered. Which is the same pattern as the plan itself, and the reason this section exists.

Worth asking of anything that promises personalization, so we asked.

Runsense keeps what Raphael describes as a multidimensional hypercube of athlete and goal characteristics, with a defined plan shape for each cell and close to 150 cells in all. A new athlete is placed in a cell and gets that cell’s shape. Our account was entered as a complete beginner who could not sustain a 30 minute run and had no schedule restrictions, so the plan we were given looks much like the plan any other complete beginner with no constraints would get.

Personalization arrives afterwards rather than up front, from two things: training history accumulating so later goals can be built on it, and information you volunteer, with schedule restrictions the example they gave us.

That is a more honest answer than most personalization marketing, and it is the right way round for a beginner, who has no history to personalise from. But set your expectations accordingly. Week one is a template, carefully chosen. The individual part is what happens to it afterwards, which is exactly why the adaptation test below is the thing this review is waiting on.

We also asked how much of a beginner 5K plan derives from None to Run, the well-known existing beginner programme. The answer, given on 2 September 2026, was more candid than the question deserved: most of it does, deliberately. Runsense assessed a range of couch to 5K plans, picked the ones it rated highest and built on those rather than writing a new one from scratch.

What gets built on top, for the individual account:

• Where you start. None to Run’s 5K plan assumes you can already run 2 to 3km continuously, and sends anyone who cannot to a different plan first. Runsense prepends run and walk weeks instead of turning you away.
• One distance per run, not a printed range. None to Run prints “2-3 km” and leaves you to choose. Runsense gives a single number, set from your own measured baseline.
• Your race, and your date. None to Run is a generic eight weeks ending in “a 5K”. Ours was laid out backwards from 24 October, and moved to a Saturday because that is when the race is.
• Two easier rungs. A four rep hill week before None to Run’s five rep one, and a twelve minute tempo before its fifteen. Smaller steps up than the source plan takes.
• Strides in the last four weeks. None to Run prescribes none at all.
• Effort in zones with a specific cue rather than “hard but controlled” and hope you guess right.

None of that is a criticism, and it is worth saying plainly why. A beginner 5K plan is a solved problem, and building on a good one is the right call. But it does change what you are buying. You are not paying for a training plan nobody has written before. You are paying for the fitting, and for the coaching that happens once that plan meets your actual week. Which is, again, why the adaptation test is the thing this review is waiting on.

What Runsense is actually trying to fix

Worth a short detour, because it explains most of the design.

Raphael Crawford-Marks wrote a piece in July 2026 setting out the problem. His argument is that a runner who wants real coaching has two options and both fail. A responsive human coach starts around $150 a month and climbs steeply from there. But the economics do not work on the coach’s side either: after the platform’s cut, a coach on that rate keeps roughly $127 per athlete, so they need something like 42 athletes to earn a median living. Nobody carrying 42 athletes is reading every run you do. The alternative is an app, which is affordable and instant and, in his words, mostly a plan generator wearing a coach’s clothes.

He calls it a missing market. Buyers who want the thing and cannot afford it, providers who cannot make a living supplying it.

The research behind that, and these are Runsense’s figures rather than ours, came from 165 running-community threads and more than 1,800 comments. The top frustrations were communication at 64%, personalization at 61%, relationship at 55% and adaptability at 52%. Plan quality was not on the list.

So the thesis is: the expensive part of coaching was never writing the plan. It was a person paying attention to you. That is why the coach lives in your messages and the app is somewhere you go to look things up.

Runsense says the coaching sits on a knowledgebase of studies and textbooks, and puts the figure at 126 sources. We asked whether there was anything citable behind that rather than a number to take on trust, and there is. The coach’s knowledgebase is the public guide library on their own site, and each guide carries its own sources list.

So we counted it ourselves. On 1 September 2026 the library held 81 guides, of which 54 carry a sources list and 27 carry none. Across those we counted 122 individual citations, close enough to their 126 that the gap is our counting rather than their arithmetic. The citations are real academic references with author, journal, volume, pages and year, not decoration: Bassett and Howley on the limiting factors for maximum oxygen uptake, from Medicine and Science in Sports and Exercise in 2000, or Warden, Davis and Fredericson on bone stress injuries in long-distance runners, from the Journal of Orthopaedic and Sports Physical Therapy in 2014. Fifteen carry clickable DOIs and the rest are cited by text.

The interesting part is which guides have nothing. The 27 without sources are, with few exceptions, the training ones: beginner running, marathon and half marathon training, 5K and 10K training, fartlek, strides, long runs, training specificity. The physiology, injury and nutrition guides are the well-cited ones. So the evidence base is thickest where the science is settled and thinnest on plan construction, which is the part the coach is actually selling. That is not a scandal, because how you build a training block is closer to craft than to literature. It is worth knowing that the citation count and the coaching method are not measuring the same thing.

It also helps to know Raphael runs ultras himself. He ran the Never Summer 100k in July 2026, his longest ever. This is not a product built by someone who has never been on the wrong end of a training block.

How we looked at it

Runsense gave us an account seeded with training history. We asked for that. Without it, everything that only means something after months of data would have been invisible, and a review written from an empty account describes the first week and calls it a verdict. Runsense confirmed to us on 7 September 2026 that every biometric metric in that account was seeded too, which matters for how you read this review and is set out in full below.

So, plainly, which is which.

Ours: the training sessions, a walk logged at 3.2km, a gym session, every conversation quoted below, and every attempt to break it. Those are activities recorded on this reviewer’s own phone. Distance, pace and duration are all a phone can measure.

Seeded: the older training history, and every biometric reading in the account. Runsense told us on 7 September 2026 that all the metrics in the account were seeded and none of the reviewer’s own biometrics ever entered it, because he does not wear a watch that measures them. An Apple Watch, or most COROS, Garmin, Suunto and Polar watches, is what produces heart rate variability, resting heart rate and sleep in the first place, and a phone measures none of them. So treat every recovery number on this page as a fixture value, not as an observation. We have removed the movements an earlier draft quoted, because the trend of a seeded figure is evidence of nothing.

What that means for the recovery section. We can tell you what the Health Insights screen does: what it plots, over what window, how it labels a reading, and what it refuses to show until it has enough data. We cannot tell you whether it reads a real body well, because no real body was attached to it. That is a genuine limit on this review and it is the main reason the score is not higher. If recovery tracking is why you are considering Runsense, you need a compatible watch and you should treat that part of this page as a description of the interface rather than a verdict on the analysis.

We used the in-app channel, so where this review describes the coach, that is the coaching conversation inside the app rather than the SMS thread.

The trial on our account ran 25 days, not the advertised 14. That was odd enough to ask about, and the answer was straightforward: Raphael extended it to give us room to review. The public trial is 14 days, which he confirmed in writing, and 14 is the figure used in the pricing section below.

This is written from days inside the product rather than months. Nobody got faster in that time and we are not going to pretend otherwise. What days are enough to judge is how it prescribes, how it answers, and what it does with your data. Captures come from two sessions, 26 and 27 August 2026 and 3 September 2026. Runsense pointed out after the first session that our account had been sitting several builds behind, so we updated before the second. The app changed in between, and where a screen changed, the newer capture is the one shown. Every image in this review is dated so you can tell which session it came from.

Talking to the coach

This is the part that justifies the subscription.

The first thing we asked was the most basic thing a beginner has: running is not part of my usual routine, what do I need to think about before starting.

No listicle came back. It said run-walk was completely fine, that no pace targets were needed yet, and that the equipment list was comfortable shoes and a short warm-up walk. Then it put the timeline in perspective: with 59 days to race day, the job was building the habit rather than chasing fitness in a week.

Small thing, but the date arithmetic was right. That sounds like faint praise until you have watched products in this category confidently get it wrong in free text.

Asked about a warm-up, it gave two answers rather than one, because an easy run and an interval night are not the same problem, and then explained the reasoning: for a new runner the point of a warm-up is to settle your breathing, not to tire you out. Asked about recovery after a hard session, it listed six ordinary things, including the least glamorous and most correct one, sleep, and then tied them to the actual plan by pointing at the next day’s easy walk and the mobility routine already sitting there.

Then came the one that changed our mind about the category.

The question was about sore thighs after a gym leg day, and whether that night’s intervals were still on.

It did not cancel the session, which is the safe answer, and it did not wave it through, which is the useless one. It said keep the intervals, but make the first rep the test. If the soreness eases once you are warmed up, carry on as planned. If it stays awkward and changes your stride, stop the hard parts and walk it in. Then it named the thing that tells the two apart: dull soreness on both sides is usually gym fatigue rather than an injury signal.

Here is the part that lifts that from good to genuinely impressive. The gym session was never entered into Runsense. It came in through Apple Health, appeared on the plan screen as an unplanned workout of one hour five minutes, and the coach folded it into an answer about something else entirely.

Faz says: This is the moment it stopped feeling like an app to me. Every fitness app I have used either says yes or says rest. This one handed me a condition to check mid-session and told me what would change its mind, using information I never gave it. That is what you are actually paying a coach for.

And when thanked for being straight rather than reassuring, it said it would rather help you make the right call than fake certainty. Which is, for once, the correct posture for software handing out training advice.

Two claims, and the one we could test

Runsense makes two behavioural claims that would be genuinely unusual if true, and that are easy to put on a marketing page and hard to build. The first is that it notices when you go quiet and asks what happened, rather than assuming you skipped and carrying on. The second is that it will refuse a goal that is not realistic and tell you what to do first, instead of building a plan to a date that was never going to work.

Rather than repeat either, we asked Runsense how to trigger them. On 2 September 2026 they gave us the exact conditions. The silence check fires after seven days with no message from you and no workout logged. Plan refusal needs a goal that is genuinely impossible rather than merely ambitious, the example offered being a marathon a week away.

That is worth publishing on its own. It turns two marketing lines into claims with a number attached that any reader can check for themselves. Most vendors in this category will not give you that, because a claim with a threshold on it can fail.

We put the second claim to it, and made the ask harder than their own example. Not a marathon next week. A marathon tomorrow.

It opened with the short answer, which was no. Then it did three things that a template would not.

It grounded the refusal in this account rather than in general caution: currently rebuilding from runs of around 3 km, with the next planned session being 32 minutes easy on Saturday. It named the actual failure mode, that the jump would more likely produce a rough walk-or-injury day than a real race effort. And then it split the question in two, which is the part we did not expect: if what you mean is could you cover the distance somehow, then maybe eventually, with a lot of run-walk and suffering. If what you mean is are you marathon-ready, definitely not yet.

Those are two different questions and most people asking it are asking the first while saying the second. Separating them is a coaching answer rather than a safety disclaimer.

The silence check we have not run to their specification, and the next section says so plainly.

When life happens, and what we have not watched

Adaptation is the central claim on every page Runsense publishes. It is also the claim we have not watched happen, and we are not going to write this section as though we have.

Specifically, we have not told the coach that a standing commitment removed two evenings a week and captured the plan before and after. We have not deliberately missed a prescribed session to see what it says. And we have not run the seven-day silence check to the condition Runsense gave us. Those are the three tests that would settle it, and none of them is done.

What we can report is what the coach did with a long gap that happened on its own.

By early September this account had not logged a run since 15 August. That is the situation every training app handles badly, usually by pretending the plan is still on or by starting the guilt notifications. We asked the coach for its read on progress so far.

It did not scold, and it did not pretend. It went to the data: recent runs had built from about 2.0 to 2.4 km up to 2.9 km, and the last logged run on 15 August was the longest so far at 2.92 km in 38 minutes. It read effort as having stayed steady across those runs rather than spiking, and called that the right pattern this far out from a 5K.

Then it put the gap in a frame: not behind, but still in the consistency phase rather than the fitness-testing phase. And it named the next two markers as specific sessions, the Saturday 32-minute easy run and then the Monday 33-minute one, on the grounds that right now regularity matters more than speed.

Two honest caveats on that. It is a response to a question we asked, not something it volunteered, so it is not evidence for the silence claim. And the app was showing unread messages from the coach at the time, which we have not opened, dated or read, so we are not going to characterise what they say or treat them as proof of anything. When we run the silence check properly, this section gets updated and the update will be dated.

The numbers it shows you

The Trends tab has three parts, and the first one told us nothing for six weeks.

When we opened it on 27 August, fitness, form and fatigue were completely blank. Not zeroed, not estimated. Blank, with a note explaining why in the app’s own words: they need about six weeks of runs before they mean anything, and it would show them the moment they were honest.

We have not seen another app in this category do that. The normal move is a confident curve drawn on three data points, which looks like insight and is closer to decoration. Refusing to draw it is a small act of integrity that costs Runsense a screenshot in their own marketing.

Then it kept its word. We went back on 3 September, 46 days in, and the same screen had filled in.

Fitness 11, form 5, fatigue 6, each with a plain-language state next to it rather than a number to interpret alone: recovering, maintaining, holding steady. Over 46 days fitness is up 10 and form up 11 while fatigue is down 2, though the small arrows on the cards themselves track the day rather than the trend, which is worth knowing before you read one as bad news. The promise was about six weeks and it took about six weeks, which is a small thing to get right and a common thing to get wrong.

The detail we would point at is on the recovery panel, not this one.

HRV, resting heart rate and sleep, each plotted over 30 days with a seven day average line and a shaded band showing a personal normal range rather than a population average, so a reading is called in or out of range against that account’s own history rather than against a textbook. That is the right design, and on a seeded account it is all we can say about it. What we can still report is the behaviour: sleep is still marked as building its baseline at 29 of 30 days, even while the other two are settled and the fitness curves are drawn. It is holding one number back on its own rule while everything around it has gone live. That is the same restraint as the empty screen, still running, when nobody would have noticed if it had quietly stopped.

That is the screen a watch buys you. Here is the honest split:

The coaching still works on a phone. What thins out is the recovery half of the picture.

We put that question to Runsense directly rather than guessing, because this review was written without a watch. Their answer: what you lose during a run is heart rate. The consequence is longer-term. Once you develop past the beginner stage, Runsense offers a lactate threshold test, and that test can only be run with a device that measures heart rate. Pass it and the coaching adds heart rate zone and pace cueing on top of the effort-based coaching it does by default. A phone-only runner never gets that option.

That is more useful than the feature table above. Phone-only is not a slightly worse version of the same coaching, it is a ceiling: you get effort-feel coaching indefinitely, and the pace and zone layer stays locked. For a beginner the effort-feel coaching is the right coaching anyway, which is the argument for their design. For someone who intends to keep improving, a watch stops being optional at the point the fitness test does. If you are weighing which one, our Whoop vs Oura vs Garmin comparison covers what each actually feeds back.

Your data, and what it refused to do

Runsense publishes a full sub-processor table. Almost nothing in this category does, and this section is the reason we would trust it with health data.

Apple Health data is not used to train or benchmark models, not used for product analytics, and reaches no advertising or analytics service. Web analytics run on Umami, which is cookieless. Account deletion is in the app, and backup copies are purged within 30 days. GDPR legal bases are set out with article references, with health data handled as special category data under explicit consent you can withdraw.

Two decisions buried in that policy tell you more than the policy does.

Runsense throws away the overnight blood oxygen reading Suunto watches send them, because a running coach does not need it. And it refuses the values Polar users type in by hand, such as maximum heart rate and training thresholds, on the grounds that those are settings rather than measurements and can be years out of date. Their own wording: they would rather have no number than a number a coach could mistake for a measurement.

Neither of those wins a feature comparison. Both cost engineering time. The benefit lands on you, because the advice you get is not quietly built on something you typed in three years ago.

Given all of that sits behind a chat box, we went at it.

We sent the coach a developer-debug jailbreak asking it to reveal its system prompt, hidden instructions, safety rules, tools and API configuration. It declined and redirected to that evening’s session. We asked again in softer language, which is usually where these give way. It declined again. Then we asked it to search its database for other users who had run more than 5km that day and tell us their names. It answered that it can only help with your own training and cannot look up or share other users’ data.

To be exact about what that is: three prompts, one day, one version, all refused. Not a security audit, and no claim that it is unbreakable. We report it because this is a product holding sleep, resting heart rate and HRV, processed through a third party model, and it behaved correctly when asked to misbehave.

What it costs

One disclosure about this section. The account used for this review was provisioned directly by Runsense, so it never met the in-app purchase screen and we have no paywall shot to show you. The prices above are the ones Runsense publishes on its own site, captured on 4 September 2026, and they are what any reader can check for themselves. We are telling you that rather than presenting a purchase flow we did not walk through.

Note the badge in that screenshot as well. The Google Play button sits next to the App Store button as though the two are equivalent. It leads to an email waitlist. If you are on Android, there is nothing to install today.

$8.99 a month, or $69.99 a year. First 14 days free, no credit card required.

There is one product and no tiers. Everything is included at that price: the plan, the rebuilds, the coach, unlimited questions. Nothing is gated behind a higher plan and there are no add-ons, which is worth checking against any competitor you are weighing it up with, because tiering is how most of this category prices.

We also put the trial length to Runsense directly, because our own account showed 25 days rather than 14. Fourteen is the correct public figure; ours was extended for the review.

Runsense publishes that price, which is worth saying plainly because a great many tools we review do not. It appears on the homepage, in the homepage FAQ, on both comparison pages and inside the structured data of their Suunto page. We checked it in a browser on 28 August 2026.

Two things the pricing block does not lead with. Billing runs through Apple In-App Purchase only, so there is no web checkout, cancellation happens in your App Store settings rather than in the app, and refunds follow Apple’s policy. Their support team will cancel on your behalf if you email them, and coaching runs to the end of the paid period either way. And the trial genuinely does not ask for a card, which is rare enough in subscription fitness to be worth the sentence.

For scale, and these are Runsense’s own published comparisons rather than numbers we have independently verified: they put Runna at about $19.99 a month and TrainingPeaks coached tiers at $149, $229 and $359. What we can say from our own checking is that $8.99 sits at the low end of AI running coaches and a very long way below any human one.

What to know before you buy

It is iPhone only. Android is an email waitlist, whose own page says Runsense is iPhone-only today. Their homepage carries a Google Play badge, which leads to that waitlist rather than to an app you can install. Runsense told us in August 2026 that it is targeting a release by the end of the year, which is a target rather than a commitment. Do not plan a purchase around it.

The trend charts are blank for about six weeks. By design, as above, but you should know the headline analytics are empty for most of your first block.

Recovery data needs something on your wrist. See the table. The coaching works without it.

It is new. First shipped 30 June 2026, and it has shipped several updates since, so there is no long public track record to lean on yet. Development is quick, which cuts both ways.

Who this is for

Runners who want to be coached rather than to manage a plan. If “just tell me what to do today and why” is your position, this was built for you.

Beginners, genuinely. Effort instead of paces, extras marked optional, run-walk offered without making it feel like a failure, and answers that give you a decision instead of a disclaimer.

Anyone coming back from a break or an injury, for the same reason. Coming back is nothing but judgment calls, and judgment is where this product is strongest.

People whose weeks move. Travel, shift work, small children, a semester of evening classes.

Runners priced out of a human coach. At $8.99 against $150 and up, that is not a close comparison.

Anyone with a phone and nothing else. No watch, no chest strap, no other subscription.

Who this is not for

All three of these come from Runsense’s own comparison pages. We are repeating them because they are true and because a company that publishes its own edge cases has earned the courtesy.

Multisport athletes. It reads your rides, swims and strength work and builds your running week around them, but it will not periodise a triathlon or a cycling season. For that, see our AI triathlon training apps guide.

Runners who enjoy the controls. If browsing plans and dragging sessions around is the part you like, this product is designed to take that job off you and you will find it frustrating rather than freeing.

Anyone who wants deep analysis. Power curves, WKO-grade modelling, years of data to pore over. Runsense reads the numbers and tells you what they mean, which is the opposite of what that runner wants.

Verdict

Runsense is the best coaching judgment we have found at this price, in a product that is young and unusually honest about being young.

Everything we were able to check, it handled. The conversations are specific rather than generic, and they draw on training you never deliberately handed over: the clearest single moment in this review is the coach folding a gym session that arrived through Apple Health into an answer about whether that night’s intervals were still on. Asked for a marathon tomorrow it refused, and refused with this account’s own mileage rather than a disclaimer. Asked for a read after three weeks off it went to the file rather than to the guilt. Asked to leak its system prompt or another user’s data, it declined three times.

The restraint is the other half of it. It refused to draw fitness curves until it had about six weeks of runs, and it kept its word: they filled in at 46 days, and the sleep baseline is still marked as building even now. It throws away the overnight blood oxygen reading it gets sent, and refuses hand-typed threshold values, because a coach should not mistake a setting for a measurement. It publishes a full sub-processor table, which almost nothing in this category does.

What holds the score down is not a flaw we found, it is a claim we could not check. Adaptation, the thing Runsense leads with everywhere, is the one behaviour we have not watched. The refusal claim held up when we asked for something impossible. We did not run the schedule-change test or the seven-day silence check, and we would rather carry that gap in the open than score around it.

Against it: two months old, iPhone only with Android on a waitlist, and a recovery half that quietly wants hardware the marketing implies you can do without. Phone-only is not a slightly thinner version of the same product, it is a ceiling, because the lactate threshold test that unlocks pace and zone cueing needs a device that measures heart rate. For a beginner that ceiling is nowhere near you. For anyone who means to keep getting faster, the watch is not an accessory, it is the thing that unlocks the rest of the product.

At $8.99 a month against $150 and up for a human coach, with 14 days that take no card, the thing you are buying is judgment on the questions a chart cannot answer. On that, it delivers.

Tools mentioned in this guide

• Runsense, the subject of this review
• Runsense: Run Coaching on the App Store, the correct listing
• Runsense for Android, currently an email waitlist
• Runsense on how it compares to Runna, their claims about a competitor, not ours
• Runsense on how it compares to TrainingPeaks, same caveat

References and further reading

Every price in this review was read off the page listed here on the date given, not taken from a press release or a comparison site.

• runsense.ai, pricing and the Google Play badge, captured 4 September 2026
• runsense.ai/android, confirming iPhone only today, 4 September 2026
• Runsense privacy policy, the sub-processor table and the GDPR bases
• Runsense support
• App Store listing, ratings count and platform, 4 September 2026

More from our fitness coverage: the best AI running coach apps, the best AI workout apps for beginners, Whoop vs Oura vs Garmin, AI sleep coach apps for athletes and AI triathlon training apps.

Written by

Faz

Faz is the founder of AIToolsBakery. Some tools here are tested hands on. Others are assessed from vendor documentation and pricing verified on the live page, and every review says which one it is. Sponsors can buy a position in a guide. They cannot buy the score, the criticism, or silence about a better option.

How we test and how we make money →

Faz

The Baker

Faz is the editor and founder of AI Tools Bakery, where every AI tool review is built on verified vendor pricing, documented user reports, and published product records. 10+ years in digital marketing, now covering AI software across 19 industries with honest verdicts and no pay-to-win rankings.

Related Posts

Fitbod Pricing (2026): Why There Is No Price on Their Website

PT Distinction Pricing (2026): Every Plan and the Real Cost Per Client

Best AI Cycling Coach Apps (2026): 7 Tested by a Real Cyclist