Get the reviews that fill next week: 10 AI skills for your reputation

fake-review-takedown

get an invented review removed

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: fake-review-takedown
description: Builds the removal case for a review that was invented. It tests the review against the platform's own published policy clause by clause, matches each ground to evidence you actually hold, writes the report, tracks the appeal, and prepares the statutory notice of complaint for the rare case where the platform refuses and the statement is false and damaging. It refuses to report a review that is merely unfair. Use when a review describes a visit that did not happen.
---

# The invented review taken down, with the clause and the evidence that did it

You give this the review, your records for the date it claims, anything you know about the account that posted it, and the platform. You get back the exact policy clause it breaches quoted from the platform's own page, the evidence lined up against that clause, the report written in the platform's own language, the appeal if the first report fails, and a dated log. Where nothing in the policy fits, it tells you so and stops, because a report with no ground is a wasted appeal you only get one of.

## What it does

1. **Test first whether this is removable at all, because most bad reviews are not.** Google states it plainly: "You can report any review, but only those that violate Google policies are eligible for removal", and then, in the same breath, "Do not report a review just because you disagree with it or dislike it. Google doesn't get involved in conflict between businesses and customers. Negative reviews can highlight areas for improvement and aren't always a sign of poor service." A review that is harsh, wrong about the wine, or written by someone having a bad day is not a fake review. Sort the review into removable or not removable before any work begins, and if it is not removable, say so and hand it to the reply skill.

2. **Name the clause, quote it, and stop looking once you have the strongest one.** Google's prohibited content is organised into named categories and a report has to pick one. "Fake engagement" covers "Content that is not based on a real experience or does not accurately represent the location or product in question" and "Content that has been posted from multiple accounts by or at the request of one person". "Rating manipulation" covers content "based on a conflict of interest", which includes "industry competitors, familial relationships". "Impersonation", "Misrepresentation", "Personal information" and "Offensive content" each have their own wording. Tripadvisor's grounds are different again and are listed under its own headings. Quote the clause verbatim into the report. A report that describes the unfairness in your own words is answered by a machine that was looking for a category.

3. **Match every ground to a document you can open today.** No booking under that name on that date, from the booking system. No table for six that evening, from the till. The dish they complain about has not been on the menu since March, from a dated menu file. The account has reviewed four venues in one town within an hour, from the public profile. A named member of staff who was not working, from the rota. One ground with one document beats five grounds with an account of what you remember. Write each pair down: clause, evidence, where the evidence lives.

4. **Use the unglamorous grounds, because they succeed more often than the dramatic ones.** Tripadvisor will not post a review from someone who was not a guest: "If you did not experience the business as a guest or potential guest, we will not post your review", and "We will not post reviews from customers who only relay a brief phone conversation with a property or property representative." It removes "Comments solely on a property's policy", "Comments solely on business hours or closures" unless the posted information created an expectation of service, and "Unsuccessful attempts to contact a business", noting that "We do not consider unanswered calls or emails to be a valid description of the service provided by the business". It also removes reviews commenting on media coverage of an event the writer was not present at, and enforces a window: "all reviews must be submitted within one year of your experience". Several of the reviews that most annoy owners fall into one of those boxes.

5. **If you think it is a competitor, say so through the tool and not in a reply.** Tripadvisor's instruction is specific: "If you believe a review has been written by the owner or competitor of a listing on Tripadvisor, please use the reporting function on our site to let us know. Reviews that allege other posted reviews are fraudulent will not be posted as that information does not relate to your travel experience." So you may report it and you may not write about it anywhere public, including under another review. The platform also names the behaviour it is looking for, "Vandalism", meaning attempts "made by individuals affiliated with one property, to damage the reputation of a competing property", and gives a route for a pattern rather than a single review, an email address for content integrity published on the same page.

6. **Submit through the route the platform documents, and record the status it gives back.** On Google that is the profile itself, "Go to your Business Profile. Select Read reviews. Next to the review that you'd like to flag, select Report", or the Reviews Management Tool for tracking. Google says "Review evaluation typically takes several days" and returns one of three statuses: "Decision pending", "Report reviewed - no policy violation", or "Escalated - check your email for updates". Write the status and the date into your log rather than trusting memory, because the next step depends on which of the three you got.

7. **Spend the appeal carefully, because there is one.** Google's wording is "If a flagged review doesn't qualify for removal, submit a one-time appeal", made through the Reviews Management Tool, where "You can select up to 10 reviews" and the outcome arrives by email. One appeal means you do not appeal in temper the same afternoon. Go back to step 3, find the document you did not attach, and use the appeal to supply evidence rather than to restate the complaint. If several invented reviews arrived together, appeal them as one batch, because a pattern is more persuasive than a single account and the tool is built to take up to ten.

8. **If the platform refuses and the statement is genuinely false and damaging, there is a statutory route, and it belongs to your solicitor.** Section 5 of the Defamation Act 2013 gives a website operator a defence for a statement it did not post, but section 5(3) says that defence "is defeated if the claimant shows that" the poster could not be identified, "the claimant gave the operator a notice of complaint", and the operator failed to respond as the regulations require. The regulations then set what the notice must contain and what happens next: the operator has 48 hours to pass it to the poster, and the poster has until "midnight at the end of the date specified in the notification as the deadline for responding (which must be the 5th day after the day on which the notification is sent)". Where the operator "has no means of contacting the poster", it "must, within 48 hours of receiving a notice of complaint, remove the statement". This skill prepares the contents of that notice. It does not send it, and nobody should send one without advice.

9. **Log it, and know the three things that turn your problem into your offence.** Do not offer the reviewer anything to take it down, do not pressure them, and do not answer a fake review by adding reviews of your own. Paragraph 13(1) of Schedule 20 to the Digital Markets, Competition and Consumers Act 2024 bans submitting or commissioning a fake consumer review, and the CMA's guidance warns traders not to "limit access to and/or the impact of negative reviews by editing, withholding or removing such reviews". Removing a genuine negative review is the same family of conduct as inventing a positive one. Keep the log: the review, the date, the clause claimed, the evidence attached, the status, the appeal, the outcome and the date of each.

## Then it checks

1. The review has been sorted as removable or not removable, and every not removable review has been handed to the reply process with the reason recorded and no report submitted.
2. Every ground claimed is a verbatim quotation from the platform's own published policy, with the URL of that page and the date it was read.
3. Every ground is paired with at least one piece of evidence that exists as a document or a screen today, named by system and date, with no ground resting on recollection alone.
4. The report contains no accusation about a named individual, no contact details of the reviewer, no reference to any offer made to the reviewer, and no claim the evidence does not support.
5. The log records the submission date, the status returned, and whether the single appeal has been used, and no appeal has been submitted before the first report returned a status.
6. Nothing anywhere in the process offers the reviewer money, refunds, vouchers or goods, applies pressure to remove the review, or adds a review of the venue written by anyone connected to it.

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. A report submitted by the manager is signed by the manager and the owner is told it went.
- It will refuse to build a report for a review the evidence shows is genuine, however unfair, inaccurate or damaging it is. A genuine negative review removed on a false ground is the conduct the CMA describes as limiting access to negative reviews, and it wastes the one appeal on the review you might actually have needed it for.
- Never contact the reviewer to ask them to remove or change the review, and never offer them anything. Both are prohibited by the platforms' own rules and the second can amount to commissioning a fake review.
- Never write, buy or ask anyone for a positive review to dilute a bad one. That is the banned practice itself, not a defence against it.
- Never name a member of staff, a suspected competitor or any individual in a public reply about a suspected fake review. Suspicions go through the reporting tool; there is no version of this that belongs under the review.
- Never state a removal rate, a success rate for appeals, a typical turnaround or any figure about how many reports succeed. Google publishes no such figure, no reliable one exists for UK venues, and an owner who expects a number will keep reporting past the point of usefulness.
- Never claim a review has been removed until you have looked at the live listing and seen that it is gone. A status message is not the page.
- This output is a working document prepared for the owner to check against their own records, and for their solicitor to check before anything described as a notice of complaint under section 5 of the Defamation Act 2013 is sent to anybody. It sets out published platform policies and published law and lines up the owner's own evidence against them; it is not legal advice, it is not a defamation assessment, and it is not a prediction that a review will be removed.

## Built from
- Google, "Report inappropriate reviews on your Business Profile", Google Business Profile Help, https://support.google.com/business/answer/4596773, no publication date shown on the page, read 16 September 2026: the eligibility warning in step 1, the reporting route and the three statuses in step 6, and the one time appeal with its ten review limit in step 7.
- Google, "Prohibited and restricted content", Maps User Contributed Content Policy, https://support.google.com/contributionpolicy/answer/7400114, no publication date shown on the page, read 16 September 2026: the named categories quoted in step 2, including fake engagement, rating manipulation, conflict of interest, impersonation, misrepresentation and personal information.
- Tripadvisor, "Trust and safety review posting guidelines", https://www.tripadvisor.co.uk/Trust-lvBd3L1aU38Y.html, no publication date shown on the page, read 16 September 2026: the first hand experience requirement and the second hand phone call exclusion in step 4, the ineligible experience categories and the one year window in step 4, and the competitor reporting instruction and vandalism definition in step 5.
- Digital Markets, Competition and Consumers Act 2024, Schedule 20, paragraph 13, https://www.legislation.gov.uk/ukpga/2024/13/schedule/20/paragraph/13/enacted, Act of Parliament given Royal Assent in 2024, read 16 September 2026: the ban on submitting or commissioning a fake consumer review, which is the boundary in step 9 and in the rules.
- Competition and Markets Authority, "Fake reviews" (CMA208), https://assets.publishing.service.gov.uk/media/67eeb64fe9c76fa33048c790/CMA208_-_Fake_reviews_guidance.pdf, published 4 April 2025, read 16 September 2026: paragraph 4.4, including the distinction that suppressing fake negative reviews is important while suppressing genuine ones is problematic, which is why step 1 sorts before it reports.
- Defamation Act 2013, section 5, https://www.legislation.gov.uk/ukpga/2013/26/section/5/enacted, Act of Parliament 2013, read 16 September 2026: the website operator defence, how it is defeated, and the contents a notice of complaint must have, in step 8.
- The Defamation (Operators of Websites) Regulations 2013, regulation 2 and the Schedule, https://www.legislation.gov.uk/uksi/2013/3028/regulation/2/made and https://www.legislation.gov.uk/uksi/2013/3028/schedule/made, SI 2013/3028, read 16 September 2026: the additional information a notice must specify and the 48 hour and fifth day timings quoted in step 8.

Prompt for Codex

# fake-review-takedown

## You are given
The reviews the owner believes are invented, exported from the platform, each with the review identifier, the platform, the posting date, the rating, the reviewer display name, the reviewer profile URL where public, the full review text and the date claimed for the visit if the review states one. The sort already made by Claude for each review: `removable` or `not removable`, with the reason. For each `removable` review, the ground or grounds claimed, each supplied as a verbatim quotation from the platform's own policy page together with that page's URL and the date it was read. The owner's evidence list: one row per document, saying what the document is, which system it came from, its date, and what it shows. The venue's records for each claimed visit date: bookings, covers, till transactions, rota, menu in force. Any previous report log with submission dates, statuses and appeal use. For any review going to the statutory route, the complainant's name, an email address for contact, the meaning attributed to the statement, which aspects are said to be factually inaccurate and which are said to be opinions unsupported by fact, and the owner's decision on whether their name and email may be passed to the poster.

## Produce
Write into a `./fake-review-takedown-output/` folder:

1. `reviews.csv` with these columns in this order: `review_ref`, `platform`, `posted_date`, `claimed_visit_date`, `rating`, `reviewer_display_name`, `reviewer_profile_url`, `review_text_verbatim`, `sort`, `sort_reason`, `days_since_claimed_visit`. `review_ref` is F0001 upward. `sort` is `removable` or `not removable`. `sort_reason` is exactly one of `matches a quoted policy ground`, `genuine but negative`, `disagreement about facts only`, `outside the platform's stated window`, `no policy ground found`.
2. `grounds.csv` with columns: `review_ref`, `ground_ref`, `policy_category`, `policy_text_verbatim`, `policy_url`, `policy_read_date`, `strength`. `ground_ref` is G01 upward within each review. `strength` is exactly one of `documented`, `partly documented`, `assertion only`. Any `assertion only` row is listed in `gaps.md`.
3. `evidence-map.csv` with columns: `review_ref`, `ground_ref`, `evidence_ref`, `evidence_type`, `source_system`, `evidence_date`, `what_it_shows`, `held_today`. `evidence_type` is exactly one of `booking record`, `till transaction`, `rota`, `menu in force`, `reservation diary`, `door or CCTV log`, `public reviewer profile`, `correspondence`, `photograph`, `other`. `held_today` is `yes` or `no`, and `no` means the ground is downgraded to `assertion only` in `grounds.csv`.
4. `report-text.md` - one section per `removable` review, each containing only: the review reference, the quoted policy ground, and the evidence references. Copied verbatim from the input; no new sentences are written here. Any review with no quoted ground gets an empty section and appears in `gaps.md`.
5. `report-log.csv` with columns: `review_ref`, `submitted_date`, `route`, `submitted_by`, `status`, `status_date`, `appeal_used`, `appeal_date`, `appeal_outcome`, `live_listing_checked_date`, `review_still_visible`. `route` is exactly one of `profile flag`, `reviews management tool`, `platform content integrity email`, `other`. `status` is exactly one of `decision pending`, `report reviewed - no policy violation`, `escalated`, `removed`, `no status returned`. `appeal_used` is `yes` or `no`. Appended to any previous log, never overwritten.
6. `notice-of-complaint-fields.csv` with columns: `review_ref`, `field`, `value_verbatim`, `supplied`. One row per required field: `complainant name`, `the statement complained of`, `why it is defamatory of the complainant`, `where on the website it was posted`, `electronic mail address for contact`, `the meaning attributed to the statement`, `aspects said to be factually inaccurate`, `aspects said to be opinions not supported by fact`, `confirmation of insufficient information to bring proceedings`, `consent to release name to the poster`, `consent to release email to the poster`. `supplied` is `yes` or `no`. This file records what the owner has and has not provided. It is not a notice and is not to be sent.
7. `gaps.md` - a numbered list of: every ground with no evidence row; every `assertion only` ground; every evidence row where `held_today` is `no`; every `removable` review with no quoted policy text or no policy URL; every appeal recorded before a status was returned; every review where `review_still_visible` is `yes` more than thirty days after a `removed` status; every `notice-of-complaint-fields.csv` row where `supplied` is `no`; and every `not removable` review for which a report was nonetheless logged.

## Rules
- Codex matches, cites, maps and logs. It never decides that a review is fake, never writes a report, never writes a notice of complaint, and never drafts an argument. Every judgement arrives with the input.
- Never write a policy ground in your own words. A ground is a verbatim quotation from a named policy page with a URL and a read date, or it is not a ground.
- Never upgrade a ground's strength. A ground with no evidence row is `assertion only`, whatever the input says, and that is recorded rather than argued with.
- Never write, suggest or record any offer to a reviewer, any request that a reviewer change or remove a review, or any text of a review of the venue. All three are the banned practice in Schedule 20 to the Digital Markets, Competition and Consumers Act 2024 and none belongs in any file.
- Never build a report for a review whose sort is `not removable`. Record it, list it in `gaps.md` if one exists in the input, and write no report section for it.
- Never write a removal rate, an appeal success rate, a typical turnaround or any figure about how many reports succeed. No published figure exists.
- Never record a review as removed without a `live_listing_checked_date` and a `review_still_visible` value read from the live page. A platform status is not the page.
- Counts and date arithmetic are computed, never estimated. `days_since_claimed_visit` is computed from the supplied dates, and any review whose claimed visit date is missing is recorded as missing rather than guessed.
- Every quoted review, policy line, status and field value is copied exactly, including punctuation, capitalisation and any emoji.
- Never submit a report, never connect to a platform's management centre or API, never email a platform, and never contact a reviewer.
- Use British English, GBP and DD Month YYYY dates. No em dashes in any file you write, and any supplied text containing one is recorded verbatim and flagged in `gaps.md`.
- Every file ends with this line: this is a working document prepared for the owner to check against their own records, and for their solicitor to check before anything described as a notice of complaint under section 5 of the Defamation Act 2013 is sent. It applies supplied policy text to supplied evidence and is not legal advice.

## Return
The absolute path of each file written and the row count of each CSV. The count of reviews by sort and the count of each `sort_reason`. For every `removable` review, the policy category and the first fifteen words of the quoted ground, the number of evidence rows attached and the strength recorded. Every ground with no evidence and every evidence row not held today. The number of reports logged, the count of each status, the number of appeals used and how many reviews remain visible on the live listing with the date each listing was last checked. Every `notice-of-complaint-fields.csv` row where `supplied` is `no`, by review. Any report logged against a `not removable` review. And the `gaps.md` item count.

Built from the best public work on this

Sources for fake-review-takedown

Everything below was opened and read on 16 September 2026. Nothing is cited that could not be loaded.

1. Google, "Report inappropriate reviews on your Business Profile", Google Business Profile Help

https://support.google.com/business/answer/4596773, no publication date shown on the page, read 16 September 2026.

The operating manual for the whole job on Google, and it opens by telling owners something they do not want to hear: "You can report any review, but only those that violate Google policies are eligible for removal. Flagged reviews that violate our content policies are removed and will no longer show on Maps and Search." Then, in its own words, "Do not report a review just because you disagree with it or dislike it. Google doesn't get involved in conflict between businesses and customers. Negative reviews can highlight areas for improvement and aren't always a sign of poor service." That paragraph is step 1, and it is the reason this skill has a sort before it has a report.

The route is set out literally: "Go to your Business Profile. Select Read reviews. Next to the review that you'd like to flag, select Report. Select the reason to flag the review. Examples: 'Spam' or 'Profanity'. Select Send report." For tracking there is the Reviews Management Tool, and the page says "Review evaluation typically takes several days."

The three statuses in step 6 are quoted from the page: "Decision pending: The review is flagged, but it hasn't been evaluated yet"; "Report reviewed - no policy violation: The review was evaluated and no policy violation was found"; and "Escalated - check your email for updates: The appeal has been escalated and you'll get an email about the final decision."

The appeal is the part worth planning around, and step 7 is built on two sentences: "If a flagged review doesn't qualify for removal, submit a one-time appeal", and, in the instructions, "You can select up to 10 reviews." One appeal, up to ten reviews at a time, outcome by email.

The page also carries the signpost used in the one-star-reply skill: "If your Business Profile has been targeted by an extortion scam involving negative reviews, learn how to report the incident."

Where the skill departs: the page also explains reporting a user profile rather than a review, and third party reviews shown on Business Profiles. Neither is the ordinary case for a UK venue with one invented review, and the skill leaves both out rather than padding the process.

2. Google, "Prohibited and restricted content", Maps User Contributed Content Policy

https://support.google.com/contributionpolicy/answer/7400114, no publication date shown on the page, read 16 September 2026.

This is the list a report is judged against, and step 2 exists because a report has to pick a category from it rather than describe a grievance.

"Fake engagement" is the obvious one: "Contributions to Google Maps should reflect a genuine experience at a place or business. Fake engagement is not allowed and will be removed", covering "Content that is not based on a real experience or does not accurately represent the location or product in question", "Reviews or ratings that have been paid for, directly or in kind", and "Content that has been posted from multiple accounts by or at the request of one person". The last of those is the one that matches a burst of reviews from new accounts in one evening.

"Rating manipulation" adds the competitor ground: content "based on a conflict of interest. A conflict of interest may include current or former employment, a contractual or consultory relationship or other professional or personal affiliations that demonstrate a conflict of interest (such as industry competitors, familial relationships, etc.)". It also prohibits a user posting "content on a competitor's place or business to undermine that business' or product's reputation".

Three other categories are used less often and are worth knowing: "Off topic" content, which the policy frames as "Only post content that is based on your experience or questions about experiences at a specific location. We don't allow content which contains general, political, or social commentary or personal rants"; "Personal information", where content contains someone's details "posted without their consent"; and "Offensive content", which prohibits "Unsubstantiated allegations of unethical behaviour or criminal wrongdoing" while allowing "content that describes negative experiences in a respectful manner".

Where the skill departs: the policy runs well beyond reviews into photographs, questions and answers, hate speech and restricted goods. The skill uses the categories a venue will realistically claim and does not try to teach the whole policy, because a report is stronger for citing one clause well than five badly.

3. Tripadvisor, "Trust and safety review posting guidelines"

https://www.tripadvisor.co.uk/Trust-lvBd3L1aU38Y.html, no publication date shown on the page, read 16 September 2026.

Longer and more specific than Google's, and the source of the grounds in step 4 that owners never think to use.

On first hand experience: "The author of the review should be the person who experienced the property directly", and "If you did not experience the business as a guest or potential guest, we will not post your review." On the phone call review, which is a recognisable type: "We will not post reviews from customers who only relay a brief phone conversation with a property or property representative."

Under "Ineligible experience" sit four grounds that catch a surprising proportion of the reviews that damage a venue. "Comments solely on a property's policy... We will remove reviews that simply describe an interaction that is a disagreement on the stated policies." "Commenting on a property due to media coverage", where the writer was not present. "Comments solely on business hours or closures... unless the official information posted by the business created an expectation of service." And "Unsuccessful attempts to contact a business. We do not consider unanswered calls or emails to be a valid description of the service provided by the business."

The recency rule is a ground in itself: "all reviews must be submitted within one year of your experience."

For the competitor case, step 5 quotes the page directly: "If you believe a review has been written by the owner or competitor of a listing on Tripadvisor, please use the reporting function on our site to let us know. Reviews that allege other posted reviews are fraudulent will not be posted as that information does not relate to your travel experience." The page names the behaviour as "Vandalism", meaning attempts "made by individuals affiliated with one property, to damage the reputation of a competing property", and gives a content integrity email address for a pattern of fraud rather than a single review.

One more provision is worth an owner knowing before it happens to them. On review bombing: "users who are submitting reviews to a property due to a personal or political agenda or reviews submitted solely due to news or social media exposure of a property which do not describe first-hand experiences will not be published. If a significant amount of such content is detected on a listing, we reserve the right to temporarily suspend publishing all content for that listing."

Where the skill departs: the page's rules on soliciting reviews, incentives and staff belong to the review-ask skill and are not repeated here. Its bias list is used only for the competitor ground, not as a general screen.

4. Digital Markets, Competition and Consumers Act 2024, Schedule 20, paragraph 13

https://www.legislation.gov.uk/ukpga/2024/13/schedule/20/paragraph/13/enacted, Act of Parliament given Royal Assent in 2024, read 16 September 2026.

In this skill as a boundary rather than a tool. Paragraph 13(1) bans "Submitting, or commissioning another person to submit or write" a fake consumer review or one that conceals incentivisation. "Commissioning" is defined to include "incentivising by any means".

That matters here for two reasons. First, the instinct to bury an invented one star review under a handful of glowing ones written by friends is the banned practice, committed in response to somebody else committing it. Second, the instinct to contact the reviewer and offer them something to take it down is commissioning, because a review altered in exchange for a benefit is no longer based on genuine experience.

Where the skill departs: the Act gives a venue no right to have anything removed. It is not a takedown power and the skill never presents it as one. Removal comes from the platform's policy, or from the route in sources 6 and 7, and paragraph 13 is here to mark the two things an owner must not do while pursuing either.

5. Competition and Markets Authority, "Fake reviews" (CMA208)

https://assets.publishing.service.gov.uk/media/67eeb64fe9c76fa33048c790/CMA208_-_Fake_reviews_guidance.pdf, published 4 April 2025, read 16 September 2026.

One sentence carries this skill's central distinction, at paragraph 4.4: "While it is important to suppress fake negative reviews, suppressing genuine negative reviews is problematic." The guidance then lists what a trader must not do, including to "limit access to and/or the impact of negative reviews by editing, withholding or removing such reviews".

Read together, that is the whole ethical shape of a takedown process. Reporting an invented review is not only permitted, the CMA treats removing fake negative content as important. Reporting a genuine one on a made up ground is the offence. The sort in step 1 is not squeamishness, it is the line between the two.

Where the skill departs: the guidance's obligations sit largely on publishers, and a venue reporting a review to Google is not the publisher of that page. The skill uses the guidance to set the boundary and does not import the publisher duty at Chapter 8, which applies to a venue only where it publishes reviews on its own website.

6. Defamation Act 2013, section 5

https://www.legislation.gov.uk/ukpga/2013/26/section/5/enacted, Act of Parliament 2013, read 16 September 2026.

The statutory route in step 8, and it works in a way most owners assume backwards. Section 5(2) gives the website operator a defence: "It is a defence for the operator to show that it was not the operator who posted the statement on the website." Section 5(3) says that defence "is defeated if the claimant shows that (a) it was not possible for the claimant to identify the person who posted the statement, (b) the claimant gave the operator a notice of complaint in relation to the statement, and (c) the operator failed to respond to the notice of complaint in accordance with any provision contained in regulations."

So the notice does not order anybody to remove anything. It removes the operator's shelter if the operator ignores it, which is what gives the operator a reason to act. Section 5(4) defines the identification limb tightly: "it is possible for a claimant to 'identify' a person only if the claimant has sufficient information to bring proceedings against the person." Section 5(6) sets the minimum contents: the complainant's name, the statement and "why it is defamatory of the complainant", and where on the website it was posted.

Where the skill departs: everything about whether a statement is defamatory at all, whether the serious harm threshold in section 1 is met, and whether the poster has a defence of truth or honest opinion is a solicitor's judgement. The skill assembles the contents of a notice and stops. It does not advise that a claim exists and it does not send anything.

7. The Defamation (Operators of Websites) Regulations 2013

https://www.legislation.gov.uk/uksi/2013/3028/regulation/2/made and https://www.legislation.gov.uk/uksi/2013/3028/schedule/made, SI 2013/3028, read 16 September 2026.

The detail section 5 leaves to regulations, and the reason the notice fields list in the Codex brief has eleven rows rather than four. Regulation 2 adds to the section 5(6) contents: the notice must "specify the electronic mail address at which the complainant can be contacted", "set out the meaning which the complainant attributes to the statement referred to in the notice", "set out the aspects of the statement which the complainant believes are (i) factually inaccurate; or (ii) opinions not supported by fact", "confirm that the complainant does not have sufficient information about the poster to bring proceedings against that person", and confirm whether the complainant consents to the operator giving the poster their name and email address.

The Schedule sets the clock. The operator must within 48 hours send the poster a copy of the notice and a notification that the statement may be removed unless the poster responds "by midnight at the end of the date specified in the notification as the deadline for responding (which must be the 5th day after the day on which the notification is sent)". If the poster does not respond, the operator must remove the statement within 48 hours of that deadline. And where the operator "has no means of contacting the poster", it "must, within 48 hours of receiving a notice of complaint, remove the statement from the locations on the website which were specified in the notice of complaint".

The Regulations also require the operator to acknowledge the notice to the complainant within 48 hours, which is a useful thing for an owner to know is owed to them.

Where the skill departs: the Schedule goes on to cover what happens when the poster responds and wishes the statement to stay, and the position when a reasonable operator would consider the name or address given to be "obviously false". Those are steps taken after a notice has been sent by a solicitor, and the skill does not walk an owner through them.

Best public prompt we found for this job

The closest public artefact is the `legal-response` skill in Anthropic's `knowledge-work-plugins` repository, raw source at https://raw.githubusercontent.com/anthropics/knowledge-work-plugins/main/legal/skills/legal-response/SKILL.md. The repository has 24,123 stars, read from api.github.com on 16 September 2026. It generates responses to routine legal inquiries from templates, with escalation checks, which is the same architecture this job needs: a mostly mechanical process with a small number of situations that must be lifted out of it.

The part we took is its framing, stated at the top:

**Important**: This command assists with legal workflows but does not provide legal advice. Generated responses should be reviewed by qualified legal professionals before being sent, especially for regulated communications.

Its escalation triggers idea is the better borrowing. It checks, before generating anything, whether the situation is one that should not be answered from a template at all. That became step 1's sort and step 8's hand off, and it is why the notice of complaint fields are recorded as a table of what the owner has supplied rather than drafted as a letter.

We did not copy three things. It is template driven, and the templates are the user's own; here the words that matter are not the venue's at all, they are the platform's published policy, and step 2 requires them quoted rather than paraphrased. It assumes a legal team exists to review the output; a pub does not have one, so the skill is explicit about the single point where a solicitor is not optional. And it has no equivalent of the finite appeal. Google gives one appeal per review, which changes the economics of the whole process: the discipline is not to produce a good report quickly but to avoid spending the second chance before the evidence is assembled.

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.