Straight answers, including the uncomfortable ones.

MarkLens reads live public signal about a product and returns a calibrated verdict on whether it will sell, or why it is not selling. Each read runs fresh web searches, reasons only over what those results show, and grades the verdict by how much real evidence stood behind it. You get a score, the reasoning across three lenses, the market context, demand by city, and a short list of what to fix.

The model is the engine, not the product. Left alone, a language model will cheerfully invent a market size, flatter your idea, and sound certain on no evidence at all. What makes MarkLens different is everything built around the model. It searches first and reasons only over what the results show, it grades every verdict by evidence strength and lowers the score when the signal is thin, and it refuses to fabricate the numbers it cannot know. When the evidence is too thin to judge, it says so instead of manufacturing confidence. That refusal is the product.

Market research tools: the five categories →

From live web search, run at the moment you request the read, not from a stored dataset. Each read pulls current public sources such as reviews, discussion, listings, and coverage, and builds its reasoning only from what those results actually show. The sources are saved alongside the read so you can see what it stood on.

A forecast answers will this sell, for a product that is not live yet. A diagnosis answers why is this not selling, for a product already on the market. Diagnosis is the sharper tool: instead of a general outlook it looks for the one binding constraint holding sales back. MarkLens infers the mode from how you phrase your query, and you can switch it. Reach for diagnosis once you are live and have a real symptom to explain.

Market validation vs market diagnosis →

No, and that is a real limit rather than a feature held back. How much you could actually win depends on private numbers that only you have: your customer acquisition cost, your conversion rate, your margin, and your repeat rate. None of them are visible from outside, so any figure produced without them would be a guess wearing the costume of a number. MarkLens shows the category size when a grounded figure exists, describes the slice you would serve in words, and lists the private inputs that would shape your share. It will not multiply the category down into an invented total.

No. The score is a signal-based estimate, not a measured fact. It reflects how strong the public evidence looks at the moment of the read, graded by how much real signal stood behind it. Read it as an informed judgement under uncertainty, not a guarantee.

Because every read runs live web search and fresh synthesis, so the evidence it finds and the way it weighs that evidence shift a little between runs. A small movement in the score is normal and expected. A large swing is not, and usually means the public signal itself moved or the query was vague enough to pull different evidence each time.

Because that is the honest answer when the public signal is not there. When a read finds little grounded evidence, it lowers its confidence and its score instead of inventing certainty, and in diagnosis mode it will say plainly that it cannot see enough from outside to judge. A thin-evidence read is the tool working, not failing. It is telling you what is actually knowable right now.

No. A read reasons only over the sources it actually retrieved during the live search. It will not cite paywalled reports it could not open, and it will not dress up background knowledge as evidence. When a grounded figure is not there, it says so rather than filling the gap.

One free read, with no account and no card. You type in a product and you get the full verdict: the score and the headline judgement on whether it will sell or why it is not. The deeper layers open when you create a free account.

The full read. A free account reveals the three lenses with their evidence, the market context, demand by city, and the list of what to fix, plus one deeper drill on any card. Free accounts include one full read that renews each month.

One read is one full analysis of one product or question. Behind it is a single model pass with live web searches, returning the verdict and all of its depth. Re-opening a saved read, re-reading it, or sharing its link costs nothing. Your first drill on a read is included free, and going deeper with more drills is a paid feature.

Three per read on the free plan, ten on a single purchased read, and unlimited on Starter. The count is per read rather than per month, so every read starts fresh, and the limit is enforced on our side rather than in your browser. Answers draw only on that read, so asking is a way to go deeper into what was found, not a way to start a new search.

On the free plan your one read renews once a month, on the same day of the month you created your account. Sign up on the 12th and it renews on the 12th; sign up on the 31st and it renews on the last day of any shorter month. It follows your signup date, not the first of the calendar month, and not a rolling 30 days. If you need more before then, you can buy a single extra read or move to a paid plan.

Free is $0 and includes one read a month. Starter is $199 a month and includes 20 reads a month with unlimited drills and the full action plan. Two larger plans, Continuity at $399 a month and Pro at $999 a month, are coming soon and not yet available to buy. On the free and Starter plans you can also buy a single extra read when you need one, without subscribing.

No. The free read never asks for a card, whether you run it signed out or with a free account. You only enter payment details if you decide to subscribe to a paid plan or buy a single extra read.

Yes. If you have used your reads you can buy a single extra read on its own, with no subscription. It is $10 on the free plan. Extra reads you buy do not expire, and they are only used once your monthly allowance is gone.

Within 7 days of your first payment, and only if you have not run any reads on the paid plan, yes. We also correct billing errors and pro-rate refunds for extended service outages. We do not refund reads you have already run, or the unused part of a month when you cancel partway through. The full policy is on the Refunds page, and requests go to hello@marklens.ai.

Refunds and cancellation →

Your query is sent to Anthropic's Claude API to generate the read, and the finished read is saved to your account so you can return to it. Reads are stored in our database with encryption at rest and in transit. We never sell your data and we never share a read in a form that identifies you or your business. We do use reads, anonymised and pooled, to make MarkLens sharper for everyone, which is the same reason the benchmarks in later plans will be worth anything. Full detail is in the Privacy Policy.

Your reads are private to your account by default. The one exception is the Share read link: it creates a public URL, and anyone who has that URL can view that single read without logging in. There is no separate password on a shared link, so only share it with people you want to see it. Deleting the read makes the link stop working.

For a read you run without an account, we store the read itself and a salted, one-way hash of your IP address, never the raw IP in that record. The hash exists only to keep the free tier from being abused. As our privacy policy notes, ordinary server logs may still contain IP addresses, the same as on any website. If you later create an account, you can claim that read into your history.

Yes. You can delete an individual read, which removes it from your history, and you can delete your whole account from settings, which cancels any active subscription and removes your profile and reads. As stated in our privacy policy, we delete your personal data within 30 days of account deletion. You can also email requests to hello@marklens.ai.

That is the tool doing its job. Most products in this category are built to flatter, which is how founders end up building for a year on a green light that was never real. An honest negative read is the more useful one, because it caught the problem while it is still cheap to fix. Go straight to the what to fix section: that is the actionable part, and it turns the negative signal into the specific things to change.

Will my product sell? The honest answer →

It is a judgement under uncertainty, not a prediction you can hold us to. It tells you how the public signal looks right now and how well grounded that reading is. It cannot see your execution, your private numbers, or the future. Use it to make a sharper decision, not as a guarantee of the outcome.

Three things help. Be specific in your query, naming the product, the audience, and the price if you have one, because vague queries pull vaguer evidence. Use the drills to go deeper on any card that matters to you. And choose the right mode: ask will it sell before you launch, and why is it not selling once you are live with a real symptom to diagnose.