Produces the campaign rather than auditing one: two timing windows, a story prompt instead of a review request, a page instead of instructions, and the incentive rule that keeps the reviews up.
We have not measured this one. It is published untested, and the fair starting assumption is that a strong model asked to run a review campaign will write a decent email, suggest a couple of platforms and mention timing.
What the file adds is four specifics and one distinction that decides whether anything survives. The distinction is that a reward may be attached to leaving a review and never to a positive one, which rules out the rating screen that routes fives to a review link and everyone else to a support form. It also gives the resolution to the tension that creates: a support route offered to everybody is an addition, while a support route offered only to the unhappy is a filter, and the two look similar on a wireframe. The specifics are the two windows, roughly 48 hours after signup while the alternative they rejected is still recallable and one day after a named success event; the story prompt, three recall questions rather than a request for a review, on the argument that people cannot evaluate a product in general but can remember what Tuesday looked like; the destination page with a deep link to each review form, its verification requirement and its time in minutes, instead of instructions in an email; and the arrival cap that keeps reviews trickling rather than arriving sixty in a week.
Where it is a close neighbour. Launch runway plan names multi-platform reviews and the story-shaped ask in a line each, inside a dated twelve-week plan. This is the campaign, and it does not restate that plan's week structure.
Who it is not for. If you have fewer than about twenty customers, do this by hand and read this only for the incentive rule. If your category has one dominant review destination and you are already on it, that platform's own invitation tooling will do more of this than a hand-built campaign will.
Measuring one skill honestly costs about twenty model sessions: five runs with it, five without, on real material, each output graded alone by a session that is not told the other arm exists, against a rubric written by somebody who never saw the skill. We have not spent that on this one yet, so it ships labelled rather than ships silently.
How it would be measured. Five invented customer-base briefs, each naming a product, four candidate destinations with their verification requirements, an instrumented success event and a support-ticket state per account, with one brief deliberately supplying no success event at all. Graded mechanically on whether between two and four destinations are selected, whether both timing windows appear with the second tied to a named product event, whether the request is a story prompt containing at least two recall questions rather than a request for a review, whether any incentive wording is unconditional on rating and free of sentiment-based routing, whether every destination carries a deep link to the review form plus its verification requirement and its time in minutes, whether suppression rules exist for accounts that have already reviewed and for accounts with open tickets, and whether the uninstrumented brief is routed to a manual thirty-account pilot instead of a base-wide send. The spine is computable, so Tier A.
Most of the spine is a lookup. Destination count between two and four, both windows present with the second tied to a named event, at least two recall questions in the prompt, a deep link and a verification note per destination, suppression rules present, and incentive wording carrying no conditionality on rating. All gradeable without judgement.
The part that resists measurement is the only part that matters commercially: whether the reviews the campaign produces are specific. That needs real customers, real time and real platforms, so the fair test grades the campaign design rather than the reviews, and it should be reported as exactly that.
The honest risk is that a strong control already knows to ask after a success moment and to ask for a story. If so, the measured gap will come from the incentive branch and the destination page specification rather than from the timing.
The rule that decides pass or fail was written down before any run was executed and it does not move afterwards. It is in the method note on the hub, along with the full results table including every skill that was tested and cut.
Stated plainly, because a skill that claims everything is useful for nothing.
name and description in the file's frontmatter, so you can also invoke it by name.If one destination dominates your category, use that platform's own invitation programme. It handles verification, stays inside its own policy by construction, and removes the deep-link maintenance problem entirely. It will not get you onto the other two surfaces, which is the reason a campaign exists at all.
A competent customer success manager with a list of forty accounts and an afternoon will beat this file outright on the small-base case, because they know which customers have a story and can ask them in a message that sounds like a person. That is genuinely the right answer under roughly a hundred customers.
The model with no skill is a reasonable alternative for the writing. Ask it two things: what is the difference between rewarding a review and rewarding a positive review, and where does the customer go after they click. If it answers both cleanly, the rest of this is a checklist.
For the legal position on incentives, gating and disclosure, neither this nor any model is the alternative. Those rules vary and have been enforced, and the person to ask is a qualified adviser in the jurisdictions you sell into.
--- name: review-harvest-campaign description: Builds a repeating campaign that produces specific, policy-compliant customer reviews across several platforms instead of one generic request sent once. Covers destination selection between two and four surfaces, the rule that a reward may be for leaving a review and never for a positive one, the two timing windows including the counterintuitive one at 48 hours, the three-question story prompt that decides whether a review carries specifics, a step-by-step destination page with a deep link and verification note per platform, a four-email sequence with suppression, and an arrival cap. This skill should be used when reviews need to be collected on a schedule, when an incentive is about to be attached to a review request, or when all existing reviews sit on a single profile. --- # Review harvest campaign ## The claim this skill is built on Reviews are not produced by asking. They are produced by removing steps at a moment when the customer already has something to say. The obvious approach is a single email to the whole customer base saying that a review would mean a lot, with a link to a profile page and possibly a gift card attached. It fails in a specific and predictable shape. A small number of your happiest customers reply, all on one platform, and what they write is "great product, does exactly what it says", which persuades nobody and cannot be quoted anywhere. Some of them never finish, because the platform asked them to create an account and verify a work email and they were on a phone in a car park. And if the gift card was conditional on a good rating, everything it produced is at risk of removal, sometimes months later and all at once. Two decisions settle the outcome, and both are made before a word of the email is written. The first is what you ask for, because asking for a review produces an evaluation and asking for a story produces specifics. The second is where you send them, because an email cannot walk beside somebody through a stranger's signup flow and a page can. Everything below exists to move one of three numbers: the share of asks that produce a review, the share of reviews that contain something quotable, and the share of reviews that are still there in six months. ## Part one. Two to four destinations, chosen before anything is written Never one. Never eight. **The shape that works** is one destination where the product is actually bought or installed, one category review site whose pages appear in searches for alternatives and comparisons, and one industry or vertical directory your buyer already reads. Your own website is a recycling target, not a destination, because a testimonial you host is evidence of nothing to a sceptical buyer. **Why not one.** A profile with forty reviews and nothing anywhere else reads as managed to exactly the careful buyer you were trying to convince. It is also a concentration risk: one profile suspension, one ranking change, one platform decision, and every piece of social proof you own disappears at the same moment. **Why not eight.** Every destination has its own account creation and its own verification, and verification is where completion dies. Each additional surface reduces the share of people who finish, and completion is the binding number in this campaign, not intent. Intent is abundant. Nobody minds writing you a review. They mind the third screen. **Rank the candidates on three criteria before choosing.** - **Does your buyer read it?** Evidence, not assumption. Ask five recent customers where they looked before buying, and write down what they say rather than what you expected. - **Does it rank for the comparison queries your category generates?** Search the obvious alternatives and versus queries yourself and see which review surfaces appear. - **How hard is its verification?** Record it per destination, in the plan: work email, a screenshot of the product, a social login, an invoice, a waiting period after purchase, or nothing at all. This number decides your ordering later, because you always offer the easiest destination last in the sequence, when the remaining people are the ones who nearly did not bother. Record, next to each destination, the URL of its published policy on solicited and incentivised reviews and the date you read it. That record is what makes the campaign repeatable next quarter, and it is what stops somebody attaching an incentive in six months on the basis of what was true today. ## Part two. The incentive rule, which is the compliance spine **A reward may be for leaving a review. A reward may never be for a positive review.** Everything else in this section follows from that one line, and getting it wrong is the failure that removes the reviews after you have already spent the ask. Stated operationally, it means all of the following. - **The reward is identical regardless of what is written.** Same offer, paid the same way, to somebody who gives one star as to somebody who gives five. If the reward can be withheld on the basis of content, it was conditional, whatever the email said. - **You do not see it first.** No approval step, no "send it to us and we will post it for you", no reward that arrives only after you have read it. - **No sentiment-based routing.** A screen that asks how likely you are to recommend us and then sends high scores to a review link while sending everyone else to a support form is review gating. It manufactures a public rating that is not the rating your customers actually give you, and it is the practice that regulators and platforms have moved hardest against. **The tension this creates, and its resolution, because this is where good intentions build a gate by accident.** You genuinely should offer an unhappy customer a conversation rather than a link. The distinction that keeps it lawful and honest is additive versus diverting. - **Additive, which is fine:** everybody who lands on the page sees the review destinations, and everybody also sees a direct line to a human with a line saying that if something is broken you would rather fix it than be reviewed about it. Nobody is filtered. The customer chooses. - **Diverting, which is gating:** the page decides, based on a rating or a survey answer, which of those two things a given person is shown. The same distinction applies to satisfaction surveys. Measuring sentiment privately is a legitimate and useful activity. It stops being legitimate the moment its answer determines who is shown a review link. **Three more lines that belong in the same rule.** - **Never draft the review for them.** Offering a suggested wording produces reviews that share phrasing across different profiles, which is detectable, and which reads as fabricated even when every reviewer was real and sincere. - **Never ask anyone to remove or amend a negative review**, particularly not somebody you have rewarded. Reply to it in public instead. - **Do not ask employees, contractors, investors or family.** It is prohibited nearly everywhere and it is the first thing an investigation looks for. **On dates and variance, honestly.** As of August 2026, incentive and solicitation rules differ substantially by platform and by jurisdiction. Some destinations prohibit any incentive at all. Some permit a small token where the reviewer discloses it. Disclosure obligations travel with the reward and differ again by jurisdiction. Several regimes were tightened between 2024 and 2026. Treat any advice on this point older than a year as unreliable, including this file, and check each destination's own published policy before offering anything. **The safe default, which is also usually the right one: if you cannot verify a destination's current incentive policy before the campaign date, run with no incentive.** The two windows and the story prompt do far more work than any incentive does, and a campaign with no incentive has no incentive risk. Where you want to give something anyway, prefer a non-monetary benefit that is not a payment for the act: sending the reviewer a designed quote card of their own words for their own use, or an early look at what is being built next, offered to everybody who was asked rather than to everybody who complied. ## Part three. The two windows The window everybody uses is the wrong one. Asking at purchase gets you somebody with nothing to say yet. Asking at renewal gets you somebody thinking about money. Asking in a quarterly newsletter gets you somebody thinking about nothing. **Window one: within roughly 48 hours of signup or first setup.** This is the counterintuitive one, and what it captures is not a considered product review. It captures the buying decision. What was going wrong before, what they tried instead, what nearly stopped them buying, and what the first hour was like. Those details are the most persuasive content any review can carry and they are gone within a fortnight, replaced by a vague sense that the software is fine. Two constraints. Some destinations require a tenure or a verified purchase before they will accept a review, so window one routes only to the destinations that accept it, and otherwise routes to a story you hold and use with permission. And the prompt in window one asks only about the before-state and the alternative, because asking about results from somebody who has had the product for a day produces "seems good so far", which is worse than nothing. **Window two: one day after a first real success event.** Peak sentiment is not at purchase and not at renewal. It is immediately after the product does the thing it was bought to do, and the word immediately is doing work. One day, not one week. Define the event per product and name it in the campaign: - the first invoice actually sent to a client - the first report exported and shared - the first project completed end to end - the first month closed - the first time the thing that used to take an afternoon took ten minutes This window needs instrumentation. If you cannot fire on an event, window two collapses into "day 21 after signup", which will be materially worse, and you should say so in the plan rather than pretending the trigger exists. **Suppression, which is a field and not a memory.** - Anybody who has reviewed anywhere is suppressed from every subsequent ask, permanently, across both windows and the passive channel. - Anybody with an open support ticket at send time is suppressed. Asking somebody for a public review while their problem is unresolved gets you an accurate review of an unresolved problem. - Anybody whose ticket closed in the last seven days with a workaround rather than a fix is suppressed until it is fixed, then becomes an excellent window two candidate on the day it is. ## Part four. The prompt, which decides whether the review says anything "Would you mind leaving us a review" asks for an evaluation. Almost nobody can evaluate a product in general, so what comes back is a compliment. A compliment persuades nobody, cannot be quoted on a comparison page, and has cost you the one ask you had. **Ask for recall instead, in three questions.** 1. What were you trying to do, or what was going wrong, before you started? 2. What were you using instead, or what did you try first? 3. What is different now, and is there a number attached to it? Optionally a fourth, which is the one that makes the rest believable: what nearly stopped you buying? A review carrying one honest reservation reads as real, and readers weight it far more heavily than an unbroken five stars. **Why this works.** People cannot rate a product in the abstract, but they can remember what Tuesday morning used to look like. Recall questions return concrete detail, and concrete detail is the only thing that survives being read by a sceptic. A review that names the alternative the customer rejected also does comparison work that your own marketing is not permitted to do. **Three details in the phrasing that change the output.** - **Give a length.** "Three or four sentences is plenty" raises completion and, counterintuitively, raises specificity, because an unbounded request produces either nothing or a paragraph of general praise. - **Put the prompt first.** The questions belong in the subject line and the first line of the page. A prompt that appears after two paragraphs of gratitude is a footnote, and it will be skipped. - **Ask them to complete their profile on the destination.** Name, role and organisation change an anonymous five-star rating into a citable quote, and it is the step everyone skips because nobody mentions it. ## Part five. The destination page, not instructions in an email **The email carries one prompt and one link. Everything else lives on a page.** Instructions in an email cannot be corrected once sent, break differently in every mail client, cannot be numbered without becoming a wall of text, and abandon the customer at exactly the point they need help, which is the moment the platform's own interface appears and looks nothing like anything you described. The page contains, in this order: - **A title that carries the prompt**, not the words leave us a review. - **The three story questions**, on the page, positioned so they can be copied into whichever form the person chooses. - **One card per destination**, each carrying: a deep link straight to the review form rather than to your profile or the platform's homepage; what the platform will ask for, in plain words, such as a work email or a screenshot or a social login; how long it takes in minutes; and whether the review appears immediately or after moderation, so nobody emails you three days later asking why it is not visible. - **Numbered steps with a screenshot of the screen they will land on.** The largest single drop-off in this whole campaign is the moment somebody arrives on a stranger's interface with no idea what they are looking at. - **A direct line to a human, shown to everybody**, worded as an invitation rather than an escape hatch. Additive, not diverting, per part two. - **The incentive terms if there are any**, stated once, in the leaving-not-positive wording, on the page rather than only in the email. - **A device note.** These emails are mostly opened on a phone and several review flows are poor on mobile. Say which destinations are easier on a laptop, whether that laptop runs Windows or macOS, and give a send-this-to-myself control so the person can move it without losing the link. **Maintaining the deep links is real work and it has to be assigned.** Get each one by starting the review flow yourself on that platform and copying the URL of the form. Re-check every link once a quarter, because platforms move these paths without notice and a dead link mid-campaign converts a warm customer into an embarrassed one. ## Part six. The sequence, the cadence and the passive channel **Four emails, escalating, then stop.** 1. **Trigger day.** The story prompt, one link, under 120 words, from a named person, with a reply-to that a human reads. 2. **Plus four days.** One of the three questions only, asked as a question, same link, shorter than the first. 3. **Plus seven days.** The three-minute version. Name the single easiest destination and give one link. Nothing else on the page competes. 4. **Plus eleven days.** The close. Say that this is the last email about it, and offer the alternative for people who will not create an account anywhere: reply to this message and, with permission, it is used as a testimonial. Then stop and move them to the passive channel. A fifth email produces unsubscribes and no reviews. **The passive channel** is an in-product prompt shown only to accounts past a usage threshold, never on first run, never on an error screen, never in the middle of a task, always dismissible, and at most once per quarter per person. The same suppression rules apply. It exists because the people who never open marketing email are disproportionately the ones with the best stories. **Run it as a rolling trigger, not as a blast, and cap the arrival rate.** A month of silence followed by sixty reviews in three days is the exact burst pattern that anti-manipulation systems are built to detect, and it also reads as a campaign to a human. Steady arrival is both more credible and more durable. Set a cap in the plan, for example no more than a small handful a day for a small product, and hold sends back rather than exceeding it. **Treat the send as a bulk send.** These recipients did not ask for this email. One segment at a time, watch the complaint rate, and keep it inside whatever sending discipline you already have. ## Part seven. The decision rule Run this per account before anything is sent. - **A named success event fired in the last 24 to 48 hours, no open ticket, never reviewed.** Ask. Window two, the full three-question prompt, all chosen destinations, easiest one named last in the sequence. - **Signed up in the last 48 hours, setup complete, no open ticket, never reviewed.** Ask. Window one, prompt limited to the before-state and the alternative they rejected, routed only to destinations that accept a new customer's review, plus the reply-to-me option. - **Open ticket now, or a ticket closed in the last seven days with a workaround rather than a fix.** Do not ask. Move them to a watch list and ask on the day the fix ships, which makes them a strong window two candidate rather than a lost one. - **An incentive is proposed and at least one destination's current policy could not be verified.** Run the campaign with no incentive anywhere. Do not run mixed rules per destination in a single campaign, because the wording will end up in the wrong email. - **You cannot tell, because no success event is instrumented and support data is not joined to the customer record.** Do not send to the base. Take thirty accounts somebody in the company personally recognises as successful, ask them by hand with the same prompt and the same page, and use the completion rate and the text you get back to decide whether the instrumentation is worth building. A base-wide send against an uninstrumented list produces the generic-review outcome and spends the ask on everybody simultaneously, and the ask is the scarce resource in this campaign. Traffic is not. ## Part eight. A worked example, compressed A scheduling product for veterinary clinics. Ninety paying clinics. Four reviews, all on the marketplace where the software is installed, the most recent one fourteen months old. Nothing anywhere else. **The proposal on the table:** one email to all ninety customers offering a gift card for a five-star review. **Two things are wrong before the writing starts.** The reward is conditional on the rating, which makes everything it produces removable and the account flaggable. And a single send to all ninety spends the only ask on a group that mostly has nothing to say this week. **The redesign.** **Destinations, three.** The marketplace where the software is bought, which has verified purchase and takes about four minutes. The category review site whose comparison pages appear when clinic managers search for alternatives, which wants a work email and a screenshot, is moderated, and takes about eight minutes. A veterinary practice-management directory with a free listing, no verification, three minutes, kept for last in the sequence. Each one's policy URL and the date it was read are recorded in the plan. **Incentive, none.** Two of the three policies could not be confirmed before the campaign date, so the campaign runs without one, and reviewers are offered a designed quote card of their own words afterwards, offered to everybody who was asked. **Window one** fires 36 hours after the second staff member is invited to the clinic's account, on the reasoning that a clinic with two people in the product has finished deciding. Prompt limited to the before-state and the alternative. **Window two** fires the day after the clinic publishes a weekly rota with no manual edits, which is the product's actual promise and is already an event in the system. Full three-question prompt. **Prompt.** What did Monday morning look like before? What were you using to build the rota instead? What changed, and how many minutes did it save? Three or four sentences is plenty. **Page.** Three cards, deep links straight to each form, verification and minutes stated on each, numbered steps with a screenshot of the first screen, a direct line to the practice's account manager shown to everybody, and a note that the category site is easier on a laptop with a send-to-myself control for people reading on a phone. **Sequence.** Four emails at day 0, 4, 7 and 11, easiest destination named alone in email three, reply-to-me offered in email four. Suppression on already-reviewed and on open tickets. Arrival capped at three reviews a day, which at ninety clinics means the campaign runs for weeks rather than a weekend. **Verdict: run it, with no incentive, on the two windows, into the three destinations, as a rolling trigger.** The decision that changed most was not the email copy. It was removing the conditional gift card, which would have put every review it produced at risk of removal, and replacing the single base-wide send with two event triggers, so each clinic is asked once, on the day their rota worked, rather than on the day the marketing calendar said so. What is still undecided and goes to a qualified check before the first send: whether the quote card counts as an incentive in the jurisdictions these clinics operate in, and what disclosure, if any, has to travel with it. ## Part nine. What to do with the reviews once they exist **Reply to every review, positive and negative, within a stated number of days.** The reply is read by the next buyer more carefully than the review is, and an unanswered negative review is read as agreement. **Recycle deliberately.** A published review becomes a quote block with name, role and organisation on the comparison page, ad creative, a line in the sales one-pager, and an entry in the objection library. Ask permission once, during the campaign, and record it against the account. **Link to the source rather than only copying the text.** A quote a reader can click through to verify is worth several that they cannot. **Never edit.** You may trim with an ellipsis and you may not reorder, rephrase or tidy the grammar. A customer who finds their words improved will say so publicly, and they will be right. **Re-run it quarterly.** The suppression list grows, the destination policies get re-read and re-dated, the deep links get re-checked, and the campaign becomes a standing process rather than something somebody remembers to do before a funding round. ## Failure modes **The conditional reward.** A gift card for a five-star review. Reviews arrive, then disappear in a batch weeks later, sometimes with the profile flagged. From the outside it looks like the platform being arbitrary, and the removed reviews took the ask with them, because those customers will not be asked again. **The sentiment gate.** A rating screen routes fives to the review link and everyone else to a support form. The public rating is beautiful and untrue, the support queue quietly holds the real feedback, and if it is ever discovered the correction is public and worse than the reviews would have been. **The generic prompt.** Twelve reviews arrive saying easy to use and great support. Nothing is quotable, the sales team never uses any of them, and the campaign is recorded as a success on count. **Instructions in the email.** Click-through looks healthy and completion is a fraction of it. The drop-off is at the platform's own account-creation screen, on somebody else's domain, so it is invisible in your analytics and gets diagnosed as apathy. **Single-platform concentration.** Forty reviews on one profile and nothing anywhere else. The careful buyer reads it as managed, which is the precise opposite of what forty reviews were meant to achieve. **No suppression list.** The customers who reviewed first, meaning your best ones, get chased three more times. One of them replies to say they already did it, and one of them says so publicly. **The burst.** Nothing for eight months, then sixty reviews in a week because somebody finally ran the campaign. The pattern is the one platforms are tuned to detect, and it reads as coordinated to human readers too. **Asking during an open ticket.** The customer writes an accurate public account of an unresolved problem. It is not unfair and it cannot be removed, and it will outrank the rest of the campaign in search results. **The supplied draft.** Reviews on three different platforms share a distinctive phrase because it was suggested in the email. It is detectable, it reads as fabricated, and the fact that every reviewer was genuine does not help. **Asking before first value.** Reviews saying it seems good so far. They carry no specifics, they are visibly thin, and they lower the average rating of a profile that was otherwise fine. **The unanswered negative.** One critical review sits at the top of the profile for a year with no reply. Every prospect reads it as the last word, and answering it a year later looks worse than answering it in a week. ## What this skill does not do - It does not tell you any named platform's current rules on incentives, solicitation, verification or disclosure. Those differ by platform and jurisdiction, several were tightened between 2024 and 2026, and everything here carries an August 2026 date. Read each destination's own policy, record the date, and re-check before every run. - It does not give legal advice. Consumer-protection rules covering endorsements, testimonials, incentivised reviews and gating vary by jurisdiction and are actively enforced. A qualified adviser answers those, and the answer can change what the campaign may offer. - It does not build the instrumentation. Window two needs a real product event, and without one the campaign falls back to a date-based trigger that loses most of its advantage. - It is not ongoing reputation management. Monitoring many profiles, alerting on new reviews, routing replies and tracking rating movement over time is a separate job with separate tooling. - It cannot make a mediocre experience look good. Run against a product people are quietly unhappy with, the campaign works exactly as designed and publishes that, faster, on the surfaces your buyers check. - It does not decide where the resulting quotes go. Placing them on a page, next to a price, against a competitor's claim, is a different piece of work with its own constraints.
These skills all ask your assistant to check things against your actual codebase, your actual schema, your actual design system. Locul keeps that context current on its own, from the files you already have, on your machine. Mac and Windows, free to start.