Get booked: 10 AI skills for short lets and holiday homes
review-request
the ask that gets five stars honestly
How the two work together
Claude thinks it through. Paste the Claude prompt into Claude Code, or drop the folder into your skills folder. Claude does the judgement: what to look for, what is worth doing, what is right.
Codex gets it done. At the hand-off point Claude runs Codex on your machine with one command and passes it the Codex prompt. Codex does the mechanical part and hands the result back. Claude checks it before you see it.
No API key to set up: Claude calls the Codex you already have installed. If Codex is not installed, Claude does that half itself and tells you.
Prompt for Claude
---
name: review-request
description: Produces the review request, the repair message that comes before it, and the written policy behind both, so you get more honest five-star reviews without doing anything that became illegal in April 2025. Use when review numbers are thin, when a bad review has landed, or before setting up any automated review ask.
---
# Get more five-star reviews without buying a single one
You give this your checkout process, whatever you currently send after a stay, your last twenty reviews and anything you have ever offered a guest in return for one. You get back a one-page written review policy, the repair message that goes during the stay, the request that goes after it, the reply templates for a good review and a bad one, and a log with the columns that prove you asked everybody the same way. It is built around one fact: in the United Kingdom, since April 2025, the shortcut is not a grey area, it is a banned practice.
## What it does
1. **Read what actually became illegal, and stop doing it today.** Paragraph 13 of Schedule 20 to the Digital Markets, Competition and Consumers Act 2024 makes it a banned practice to be "Submitting, or commissioning another person to submit or write" either "a fake consumer review" or "a consumer review that conceals the fact it has been incentivised". A fake review is defined by the CMA as "a consumer review that purports to be, but is not, based on a person's genuine experience". Banned practices are automatically unfair; there is no test of whether anyone was actually misled. The CMA has been able to enforce this directly since April 2025. If your current process asks a friend, a cleaner or a family member to post, it stops this morning, not after the next season.
2. **Understand that "incentivised" does not mean cash.** The CMA's fake reviews guidance states that commissioning "is not limited to making monetary payments" and lists examples including being asked to write a review in exchange for "money", "commissions", "discounts or vouchers", "products given free of charge ('freebies')", "free stays" and "invitations to events". In a short let that covers a late checkout offered in return for a review, a bottle of wine on the table with a card asking for five stars, a discount on next year, and a free night. Each of those is commissioning, and a review obtained that way is unlawful unless the incentive is disclosed in the review itself.
3. **Take the three specific practices off the table and write down that you have.** First: offering anything at all in exchange for a review. The guidance gives as an example of commissioning "Offering an existing customer a free or discounted product in exchange for a five-star review (which is not reflective of their genuine experience)". Second: asking only the guests you think liked it. The guidance describes cherry picking as something that "might be done either through suppressing negative reviews that have been submitted or by encouraging just those who are satisfied to leave reviews", and separately warns against "arbitrarily stopping and starting review invitations". Third: trading a refund for a review change. The guidance names as commissioning "Contacting a customer who has left a negative review and offering them a refund and/or a gift card if they change their review to remove the negative commentary", and says traders should not make "an offer of dispute resolution contingent on a consumer not leaving a negative review".
4. **Write the one lawful lever into a policy: ask everyone, the same way, every time.** One page. It states who gets asked (every guest, no exceptions), when (the same interval after every checkout), through what channel, in exactly what words, and who is allowed to change the wording. The CMA is explicit that this is permitted: encouraging reviews "without predetermining the contents or sentiment expressed in the review, for example by merely emailing customers generally to ask if they wish to provide a review, is not prohibited under the banned practice." The policy is what turns "we ask everyone" from a claim into a record, and the record is the only thing that distinguishes you from a host who filters.
5. **Put the repair message before the ask, because that is where the stars actually come from.** On any stay of three nights or more, send one short message partway through asking whether anything is not right that you can fix while they are there. Then fix it. A cold radiator mended on Tuesday is a five-star review; the same radiator discovered in a review on Friday is a refund, a public paragraph and a rating you carry for two years. This is not a trick and it is not conditional on anything: you fix the problem whether or not they review, and you say nothing about reviews in that message at all. Mixing the two is precisely what creates a dispute resolution offer contingent on a review, which is the practice named in step 3.
6. **Send the ask once, the morning after checkout, at 10:00.** Not at checkout, when they are loading the car and cannot write anything. Not four days later, when the stay has blurred. The message is 60 to 90 words and contains: the guest's name, one specific detail from their actual stay so they can tell a person wrote it, one sentence saying a review helps other guests decide, one link, and the sender's name. Nothing else. One follow-up is permitted, at the same interval for every guest, and then you stop.
7. **Take out of the ask everything that predetermines the answer.** No star number. No "if you enjoyed your stay". No "we are aiming for five stars and anything less hurts us". No screening question that routes happy guests to the public site and unhappy ones to a private form, which is cherry picking with extra steps. The customer-response skill in Anthropic's public plugin repository puts the standard in one line, "Never overpromise, never mislead, never hide bad news in jargon", and that applies as much to a review request as to a complaint reply. Ask for a review, not for a good one.
8. **Treat prize draws as a question for your adviser, not as a loophole.** A footnote in the CMA's guidance says that "offering the chance to earn a reward which does not guarantee a direct benefit for the reviewer (for example, traders may encourage consumers to leave reviews by telling them that they will be entered into a prize draw) is unlikely to amount to commissioning in the context of the banned practice." Read that carefully: "unlikely" is not "does not", it is a footnote rather than a rule, and it says nothing about your booking platform's own terms, which commonly ban any inducement regardless of the law. This skill therefore does not build a prize draw into the ask. If you want one, it goes to your solicitor and to a reading of the platform's terms first, and the output records that you were told so.
9. **Reply to every review, including the bad one, within seven days.** Good review: thank them, name the specific thing they mentioned, no sales pitch. Bad review: acknowledge the fact, say what has changed since, do not argue the guest's experience and do not disclose anything about their booking that they did not put in public themselves. Where a review is genuinely fake, meaning it does not purport to describe a real stay at your property, report it through the platform's own process with the booking evidence. Do not threaten. The guidance lists interfering with reviewers "through threats of harm or legal action" as a practice traders should not engage in, and a legal threat to a reviewer is the fastest way to turn one bad review into a news story.
10. **Keep the log, because the policy is only worth what the record proves.** One row per booking: booking reference, checkout date, repair message sent yes or no, ask sent date and time, follow-up sent date and time, channel, review received yes or no, rating, replied date. No column for whether you thought they would be positive, because the moment such a column exists somebody will filter on it. Review the log monthly against one question: did every guest in the period get the same thing at the same interval, and if not, why not.
## Then it checks
1. The written policy exists, is one page, and names who is asked, when, through what channel, in what words, and who may change the wording.
2. The ask message contains no star number, no conditional phrasing such as "if you enjoyed", and no routing question that sends different guests to different places.
3. Nothing of value is offered anywhere in the process in return for a review, and the output lists every existing practice that had to be removed to make that true.
4. The log has a row for every booking in the period with an ask date, and the gap between checkout and ask is the same for every row, or the exceptions are named with reasons.
5. The repair message contains no reference to reviews and no condition attached to fixing the problem.
6. Every review in the period has a reply, and no reply discloses anything about a guest's booking that the guest did not put in public themselves.
Any check fails: name it, redo that step once. Failed twice: say what is wrong and stop.
## Rules
- Public information only.
- Never invent a fact, a number or a quote.
- Anything sent in someone's name says whose name it is.
- Never write, edit, suggest the wording of, or post a review about your own property, and never ask anyone connected to the business to do it. A review written by a person with a financial interest in the property is caught even where the stay was real, because the guidance treats a financial or commercial link as information the reader needs.
- Never suppress, delay or selectively time the ask based on how you think a guest felt. The offence is in the selection, not in the sentiment, and a log showing that the unhappy weeks were quietly skipped is evidence against you rather than for you.
- Never treat a bad review as a problem to be removed. It is a problem to be fixed, and the fix belongs in the turnover checklist or the listing, not in a message to the reviewer.
- A review request sent by email or text to a past guest may also engage the direct marketing rules. The ICO states that "You must not send marketing emails or texts to individuals without specific consent", with a limited soft opt-in exception for previous customers that requires a clear chance to opt out "both when you first collected their details and in every message you have sent". Put an opt-out line in the ask and take the classification question to your adviser.
- This output is a working document prepared for the owner's solicitor or trading standards adviser to check before it is used. It reproduces published CMA guidance and legislation; it is not legal advice, and it does not determine whether any particular practice of yours infringes.
## Built from
- Digital Markets, Competition and Consumers Act 2024, Schedule 20, paragraph 13, https://www.legislation.gov.uk/ukpga/2024/13/schedule/20, read 14 September 2026: the banned practice itself and the statutory definitions of a fake consumer review and of concealed incentivisation, which is step 1.
- Competition and Markets Authority, "Fake reviews: CMA208", https://assets.publishing.service.gov.uk/media/67eeb64fe9c76fa33048c790/CMA208_-_Fake_reviews_guidance.pdf, 4 April 2025, read 14 September 2026: the non-monetary commissioning examples in step 2, the three removed practices in step 3, the permission to ask everyone in step 4, the prize draw footnote in step 8 and the threats point in step 9.
- Competition and Markets Authority, "Unfair commercial practices: CMA207", https://assets.publishing.service.gov.uk/media/691b9bd821ef5aaa6543ee6f/Unfair_commercial_practices_CMA207_18_Nov_2025__2_.pdf, 18 November 2025, read 14 September 2026: the framing of banned practices as automatically unfair with no need to show that a consumer was misled.
- Information Commissioner's Office, "Electronic mail marketing", Guide to PECR, https://ico.org.uk/for-organisations/direct-marketing-and-privacy-and-electronic-communications/guide-to-pecr/electronic-and-telephone-marketing/electronic-mail-marketing/, no publication date shown on the page, read 14 September 2026: the consent and soft opt-in position behind the opt-out rule on the ask.
- Anthropic, `draft-response` skill, knowledge-work-plugins, https://raw.githubusercontent.com/anthropics/knowledge-work-plugins/main/customer-support/skills/draft-response/SKILL.md, no publication date shown in the file, read 14 September 2026: the honesty standard quoted in step 7 and the structure of the two reply templates in step 9.
Prompt for Codex
# review-request ## You are given A folder for one UK short let containing: a bookings export for a stated period with a booking reference and a checkout date on every row, whatever is currently sent after a stay (a saved automation, an email template, a screenshot of a card left in the property), the last twenty reviews with their platform, date, rating and text, any replies already posted, a note of anything ever offered a guest in return for a review, and a settings file naming the interval from checkout to the ask, the channel, the follow-up interval, the owner's name and the opt-out route. Some bookings have no checkout date and the review export does not link cleanly to a booking. ## Produce Write these files into `./review-request-output/`: 1. `review-policy.md` - one page stating who is asked (every guest, no exceptions), when, through what channel, in exactly what words, who may change the wording, and the one follow-up interval. 2. `messages/repair-message.txt`, `messages/review-ask.txt`, `messages/review-ask-followup.txt`, `messages/reply-positive.txt` and `messages/reply-negative.txt` - five plain text templates with variables in square brackets. `review-ask.txt` runs to 60 to 90 words and carries an opt-out line. `repair-message.txt` contains no reference to a review and no condition attached to fixing anything. 3. `review-log-GUEST-DATA.csv` - columns exactly: `booking_ref,guest_first_name,checkout_date,repair_message_due,repair_message_sent,ask_due_at,ask_sent_at,followup_due_at,followup_sent_at,channel,review_received,rating,review_date,reply_sent_date,exception_reason`. One row per booking in the period. The `_sent_at` and `_sent` columns ship empty for the owner to complete. 4. `interval-check.csv` - columns exactly: `booking_ref,checkout_date,policy_interval_hours,ask_due_at,interval_hours_actual,matches_policy,variance_hours,reason`. Plus summary rows `BOOKINGS IN PERIOD`, `BOOKINGS WITH AN ASK DUE` and `BOOKINGS WITH NO ASK DUE`. 5. `practices-removed.csv` - columns exactly: `practice_found,source_file,category,why_it_goes,what_replaces_it`. `category` is `incentive`, `cherry_picking`, `refund_for_review_change`, `self_authored` or `threat`. 6. `ask-wording-scan.csv` - columns exactly: `banned_pattern,occurrences,file,line_number`. Patterns scanned: any star number, five star, 5 star, great review, good review, positive review, if you enjoyed, if you were happy, anything less, discount, voucher, free night, late checkout, upgrade, prize draw, and any routing question that sends different guests to different destinations. 7. `reply-coverage.csv` - columns exactly: `review_id,platform,review_date,rating,reply_present,reply_date,days_to_reply,discloses_booking_detail`. `discloses_booking_detail` must be `no` on every row of every drafted reply. 8. `exceptions.csv` - columns exactly: `booking_ref_or_field,problem,what_it_blocks,source_file`. ## Rules - Write drafts to files only. Never send a message, never post a review reply to Airbnb, Booking.com, Vrbo, Google or anywhere else, never open a platform or review API, and never report, flag or respond to a review on a platform. The owner sends and posts. - Never write, edit, suggest the wording of, or draft a review of the property. This produces requests and replies, never reviews. - Never draft anything that offers anything of value in return for a review: no money, discount, voucher, freebie, free night, upgrade, late checkout or prize draw. Every such practice found in the inputs goes in `practices-removed.csv` and is not carried into any output. - Never vary the ask by expected sentiment. `ask_due_at` is computed as the same interval from checkout for every row. Any row where `matches_policy` is `no` must carry a reason that is a fact about the booking, never a guess about the guest. `review-log-GUEST-DATA.csv` has no column for expected sentiment and none may be added. - Never draft a message that trades a refund, a repair or a resolution for a review or for the removal of one, and never draft a threat of any kind to a reviewer. - `ask-wording-scan.csv` must end with zero occurrences across the five message files. Any hit is removed before the file ships and recorded in `exceptions.csv`. - Never write anything about a guest's booking into a drafted public reply that the guest did not put in public themselves. Check every negative reply draft against the review text and set `discloses_booking_detail` from that comparison. - Every date, rating and interval must trace to a supplied file. A booking with no checkout date gets an empty `ask_due_at`, a reason, and a row in `exceptions.csv`. Never estimate a checkout date from a booking date. - Every message is sent in the owner's name, named in the message, so it is clear whose name it is sent in, and the ask names the sender and carries the opt-out route from the settings. - Guest first names appear only in `review-log-GUEST-DATA.csv`. Never copy a guest name, contact detail, address or booking reference into the message templates, `interval-check.csv`, `reply-coverage.csv` or any summary. - British English. £ for money. Dates DD Month YYYY. No em dash characters. No emoji. CSVs UTF-8, header row, comma separated, fields containing commas quoted. ## Return The file paths written, the bookings in the period and the number with an ask due, the count of rows where `matches_policy` is `no` with the reasons, every row in `practices-removed.csv` by category, the occurrence count in `ask-wording-scan.csv` for the shipped files, the reviews in the period and how many have no reply, and every booking held for a missing checkout date. State that nothing has been sent or posted, that no review has been written or edited, and that the process is a working document for the owner's solicitor or trading standards adviser to check before it is used.
Built from the best public work on this
Sources for review-request
Everything below was opened and read on 14 September 2026. Nothing is cited that could not be loaded.
1. Digital Markets, Competition and Consumers Act 2024, Schedule 20, paragraph 13
https://www.legislation.gov.uk/ukpga/2024/13/schedule/20, read 14 September 2026.
The primary legislation. Paragraph 13(1) makes it a banned practice to be "Submitting, or commissioning another person to submit or write - (a) a fake consumer review, or (b) a consumer review that conceals the fact it has been incentivised." Paragraph 13(2) adds "Publishing consumer reviews, or consumer review information, in a misleading way", and 13(3) adds publishing them "without taking such reasonable and proportionate steps as are necessary" to prevent and remove fake and concealed incentivised reviews. The definitions in 13(5) are what make the reach of this so much wider than owners expect. A consumer review is "a review of a product, a trader or any other matter relevant to a transactional decision". A fake consumer review is "a consumer review that purports to be, but is not, based on a person's genuine experience". To submit is to "supply it with a view to publication", and to commission is "incentivising by any means". Two words in those definitions do the damage. "Any means" removes the assumption that paying for reviews is the only offence. "Purports to be, but is not, based on a person's genuine experience" catches the review a host writes in a friend's name about a stay that did happen, which most people assume is a grey area and which is not. Step 1 of the skill is those clauses, stated in the order an owner needs them. Where the skill departs from the source: legislation states the prohibition and says nothing about how to get reviews lawfully, which is the actual problem an owner has. The skill takes the boundary from the statute and builds the method from the CMA guidance below, and it never tells an owner whether a specific past practice of theirs infringed, because that is a question for their solicitor.
2. Competition and Markets Authority, "Fake reviews: CMA208"
https://assets.publishing.service.gov.uk/media/67eeb64fe9c76fa33048c790/CMA208_-_Fake_reviews_guidance.pdf, dated 4 April 2025 on the cover, read 14 September 2026. 29 pages.
This is the source the whole skill rests on, and it is far more useful than its title suggests, because most of it is about practices that are not fake reviews at all. Paragraph 2.10 states that commissioning "is not limited to making monetary payments" and gives examples of being asked to write a review in exchange for "money", "commissions", "discounts or vouchers", "leases or loans free of charge or on more favourable terms than those offered to the general public", "products given free of charge ('freebies')", "free stays" and "invitations to events". "Free stays" is in the list by name, which settles the most common short-let practice in one phrase. Chapter 3's examples of commissioning include "Offering an existing customer a free or discounted product in exchange for a five-star review (which is not reflective of their genuine experience)" and "Contacting a customer who has left a negative review and offering them a refund and/or a gift card if they change their review to remove the negative commentary". Paragraph 4.4 says traders should not interfere with the willingness to leave negative reviews "through threats of harm or legal action", "by preventing bona fide users from leaving reviews", "by arbitrarily stopping and starting review invitations" or "by making an offer of dispute resolution contingent on a consumer not leaving a negative review". Paragraph 4.5 defines cherry picking as something that "might be done either through suppressing negative reviews that have been submitted or by encouraging just those who are satisfied to leave reviews". And paragraph 3.6, which is the door out of all of this, says that encouraging reviews "without predetermining the contents or sentiment expressed in the review, for example by merely emailing customers generally to ask if they wish to provide a review, is not prohibited under the banned practice." Paragraph 3.7 confirms that incentivised reviews are permitted at all only if the trader tells consumers the review has been incentivised and the review still reflects the genuine experience. Every numbered step in this skill comes from one of those paragraphs. Where the skill departs from the source, and this is deliberate: paragraph 3.7 permits incentivised reviews with disclosure, and the skill declines to build that route. The disclosure burden sits partly on the reviewer, who has no reason to carry it correctly; most booking platforms ban inducements in their own terms whatever the law says; and a review labelled as incentivised is worth very little to the reader it was written for. So the skill says the lawful route exists, says why it is not worth taking, and does not provide the wording for it.
3. Competition and Markets Authority, "Unfair commercial practices: CMA207"
https://assets.publishing.service.gov.uk/media/691b9bd821ef5aaa6543ee6f/Unfair_commercial_practices_CMA207_18_Nov_2025__2_.pdf, dated 18 November 2025, read 14 September 2026.
The wider guidance on the DMCC Act's unfair commercial practices provisions, in which the reviews prohibition sits as one of the banned practices in Schedule 20. Its value here is a single point of framing that changes how seriously an owner takes the rest. Banned practices are, in the CMA's words, behaviour that "is automatically unfair and illegal", which means there is no defence built on the idea that no guest was actually deceived or that everybody does it. By contrast the misleading actions and omissions chapters, which cover most consumer-law risk, are prohibited only "if the average consumer is likely to take a different decision as a result". Reviews sit in the harsher category. CMA208 carries a similar statement at paragraph 1.1, but CMA207 is where the two regimes are set side by side, which is what makes the distinction legible. The skill uses this in step 1 and nowhere else, on purpose: it is the reason the skill says stop today rather than review at the end of the season. Where the skill departs from the source: CMA207 runs to 64 pages across price transparency, aggressive practices, invitations to purchase and much else, and none of that is in scope here. The skill takes one framing sentence and leaves the document alone.
4. Information Commissioner's Office, "Electronic mail marketing", Guide to PECR
https://ico.org.uk/for-organisations/direct-marketing-and-privacy-and-electronic-communications/guide-to-pecr/electronic-and-telephone-marketing/electronic-mail-marketing/, no publication date shown on the page, read 14 September 2026. The page carries a notice that "Due to changes made by the Data (Use and Access) Act, this guidance is under review and may be subject to change."
Included because the review ask is a message sent to a past customer, and the channel has its own rules that sit entirely outside the CMA's. The ICO states "You must not send marketing emails or texts to individuals without specific consent. There is a limited exception for your own previous customers, often called the 'soft opt-in'", and sets the exception as an existing customer who bought a similar product or service "and you gave them a simple way to opt out both when you first collected their details and in every message you have sent". It also confirms that "You must not disguise or conceal your identity, and you must provide a valid contact address so they can opt out or unsubscribe", and that the same rule applies to texts, voicemails and "direct messages via social media". Whether a review request is marketing is genuinely arguable, since it promotes nothing and asks for nothing to be bought. The skill does not resolve that argument. It requires an opt-out line and a named sender on the ask, which costs one line and is correct on either reading, and it sends the classification to the owner's adviser. Where the skill departs from the source: the ICO page is about marketing and the skill applies its mechanics to a message that may not be marketing at all. That is a deliberately cautious reading and the skill says so rather than implying the ICO has ruled on review requests.
Best public prompt we found for this job
The closest public artefact is the `draft-response` skill in Anthropic's `knowledge-work-plugins` repository, at https://raw.githubusercontent.com/anthropics/knowledge-work-plugins/main/customer-support/skills/draft-response/SKILL.md. The repository has 24,015 stars, read from api.github.com on 14 September 2026. It drafts customer-facing replies across questions, escalations, bad news and billing errors, and it carries the standard that a review process most often fails:
**Be honest**: Never overpromise, never mislead, never hide bad news in jargon
That line is quoted in step 7, because the dishonesty in a bad review request is subtle rather than gross. "If you enjoyed your stay, would you leave a review" is not a lie, and it is a filter. The structure of its bad-news section, which tells the writer to "Be direct, don't bury the news" and to say so honestly and commit to a timeline when you do not know an answer, is also the shape of the reply template for a poor review in step 9. What we did not copy is the account-context research it front-loads, which for a holiday let means going through a guest's booking history before answering, and that is both unnecessary and, for a review reply written in public, a privacy risk. We also dropped its channel and stakeholder matrix entirely. There is one channel here and one stakeholder, and the thing that actually determines the outcome is not the phrasing of the reply but whether the owner asked every single guest the same way. Nothing in any public prompt we found addresses that, because nothing in any public prompt we found is written for the United Kingdom after April 2025.
Want this running in your business?
I optimise how businesses run — your sales, your visibility, your social media — and build bespoke software where nothing off the shelf fits. The first conversation is free. Work starts from £150 a day.
Foxera