site-copy

---
name: site-copy
description: Rewrites the words on a UK hospitality website so they sound like the actual place, replacing stock phrases with facts somebody can check, cutting every claim that has no evidence behind it under the advertising code, checking the reading level, and leaving a claims register that says which document supports which sentence. Use when your site reads like every other site, when you cannot remember who wrote it, or before you send anyone a link.
---

# Words that sound like the place, not like a template

You give this every page of your website as it stands, plus half an hour with whoever actually runs the room. You get back the copy rewritten page by page, a struck list of every sentence that was removed with the reason, a claims register pairing each factual statement with the document that evidences it, and a readability check. It will not invent a story about your building, it will not write a claim you cannot prove, and it will not sign off anything as advertising-compliant, because that is a judgement for the Advertising Standards Authority and not for a document.

## What it does

1. **Get the raw material out of a person before you touch a keyboard.** Sit with the owner, the head chef or the longest serving member of the floor team and ask four questions: what do people order twice, what do regulars complain about, what do you do that the place down the road does not, and what does somebody say when they walk in for the first time. Write their answers down in their own words, including the odd ones. Every genuinely good line on a venue website started as something somebody said out loud, and every bad one was written by a person who had never been in the building. This transcript is the source material for the whole job, and copy written without it will be true of any restaurant, which is another way of saying it is true of none.

2. **Strike the template sentences by name, so they cannot come back.** Make a list of the stock phrasing already on the site and mark each one for removal: nestled in the heart of, a warm welcome awaits, passionate about food, the finest locally sourced ingredients, something for everyone, a dining experience, we pride ourselves on. These are not bad writing because they are clichés, they are bad writing because they carry no information: a reader who has read the sentence knows nothing new. Record each struck sentence with the page it came from, so that when the site is updated in two years nobody quietly puts it back because the page looked short.

3. **Put the answer in the first sentence of every page, and write in sentence case throughout.** A person arriving on your menu page wants to know what you serve and roughly what it costs. A person on your contact page wants the address. Lead with it. The GOV.UK style guide, which is written for pages read at speed by people who did not come for the prose, uses sentence case "even in page titles and service names", with proper nouns as the exception, and it is right for the same reason: capitalised phrases slow a reader down. Then cut the introduction. An opening paragraph that sets the scene before the information arrives is a paragraph that gets scrolled past, and on a phone it is the whole first screen.

4. **Write plainly, then check it rather than assuming it.** The Home Office User-Centred Design Manual sets the target most public bodies work to: "Usually we recommend writing for a maximum reading age of 9, even if you are writing for a specialist audience." That is not an insult to your customers, it is an acknowledgement that people read a restaurant website one handed, half distracted, often in a second language, sometimes with dyslexia, and often in a hurry. As the same guidance puts it, "Simple language doesn't mean dumbing down - short, clear sentences are easier to understand for everyone." Run the finished copy through a readability tool, record the score and the tool used, and rewrite the sentences that pull it up rather than averaging the problem away.

5. **Replace every adjective with a fact, and see whether the sentence survives.** "Locally sourced" becomes the name of the farm and the distance. "Award winning" becomes the award, the category and the year. "Fresh bread" becomes what time it is baked. "Extensive wine list" becomes the number of bottles and where most of them come from. This is a writing technique and it is also a legal filter, because the CAP Code rule 3.7 requires that "Before distributing or submitting a marketing communication for publication, marketers must hold documentary evidence to prove claims that consumers are likely to regard as objective". A fact you can name usually has a document behind it. An adjective you cannot convert into a fact usually does not.

6. **Cut the claims that cannot be evidenced, and name the rule that cut them.** The CAP Code is the standard applied to marketing copy in the UK, including a website. Rule 3.1: "Marketing communications must not materially mislead or be likely to do so." Rule 3.3: they "must not omit material information or information required to be included by law". Rule 3.11 covers exaggerating "the capability or performance of a product". Rule 3.2 is the counterweight that keeps the writing human, allowing "Obvious exaggerations ('puffery') and claims that the average consumer ... is unlikely to take literally ... provided they do not materially mislead". So "the best pie in the universe" is allowed and "the best pie in town" is not, because the second is checkable and somebody will check it. Struck claims go on the list with the rule number attached.

7. **Write the allergen and dietary wording carefully, because this is the one place vague copy does harm.** The Food Standards Agency guidance for food businesses, last updated 17 July 2026, names the fourteen allergens as "celery; cereals containing gluten; crustaceans; eggs; fish; lupin; milk; molluscs; mustard; peanuts; sesame; soybeans; sulphur dioxide and sulphites; tree nuts", and for food served in a restaurant or café it says "you must supply allergen information for every item that contains any of the 14 allergens". It permits a signposted route, "by displaying a sign directing customers to ask staff for allergen information", provided the information is "visible, clearly legible and easily accessible to the customer". So the website either carries the allergen information or carries a clear, prominent line telling people to ask, with the phone number next to it. A blanket "we cannot guarantee" buried in the footer is neither.

8. **Fix the mechanical style once and apply it everywhere, because inconsistency reads as carelessness.** Use the GOV.UK conventions, which are free, published and settled. Time and date ranges take "to" rather than a hyphen: "10am to 11am (not 10-11am)". Telephone numbers "separate out the area, freephone, mobile or international code from the rest of the number". Numbers are written in numerals with commas for clarity above a thousand. Link text describes where the link goes rather than saying click here, and sits at the end of a sentence rather than in the middle where it pulls the eye off the line. Write these decisions into a one page style note and hand it to whoever updates the site next, or the drift starts on the first new page.

9. **Read it aloud to somebody who works there and was not involved in writing it, then register every claim.** If the reader stumbles, the sentence is wrong, and it is wrong even if it is grammatical. If they say "we don't really do that", the sentence is worse than wrong. Then build the register: one row per factual claim, the page, the sentence, the type of claim, the document that evidences it and that document's date. Anything with no document is struck or rewritten, not published with a caveat. Put a review date on the register, because the farm changes, the award lapses, the chef leaves, and the sentence about the chef stays on the website for three years after they have gone.

## Then it checks

1. Every page's first sentence answers the question that brought the reader to that page, with no scene setting paragraph above it.
2. Every struck sentence is recorded with its page, the reason it was struck and, where a rule struck it, the CAP Code rule number, and none of the named template phrases survives anywhere on the site.
3. Every factual claim in the final copy appears in the claims register with a named evidence document and that document's date, and no claim is published with a hedge in place of evidence.
4. The readability check names the tool used, the score reached and the date, and every sentence flagged as hardest to read was either rewritten or recorded with a reason for keeping it.
5. The allergen wording either carries the information or carries a prominent, specific instruction to ask, with a phone number beside it, and it is not placed in the footer or behind a collapsed heading.
6. Dates, time ranges, telephone numbers, numerals and link text follow one written style note across every page, and the style note exists as a file.

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. Copy written in the chef's voice, or a quote attributed to a member of staff, goes to that person for approval before it is published.
- Never write a claim the venue cannot evidence today, and refuse to write one on the promise that the evidence will be found later. The CAP Code requires the documentary evidence to be held before publication, not after a complaint.
- Never invent history, provenance or a founding story. A building's age, a family's connection to it, a recipe's origin and a supplier's location are facts with records behind them, and a charming invented one is a claim that a local historian, a supplier or a former owner will eventually correct in public.
- Never state a customer number, a rating, a number of covers served, an award or a press mention without the source and its date in the register. A figure that was true in 2023 sitting on a page in 2026 is a claim about today.
- Never write "we cannot guarantee" or "may contain" as the whole of the allergen answer. It tells a customer with a serious allergy nothing they can act on, and it transfers a decision to them that they do not have the information to make.
- Never leave a named member of staff in the copy after they have left, and never leave a supplier named after they have been changed. Both are in the register with a review date for exactly this reason.
- This output is a working document prepared for the owner to check against their own menu, prices, suppliers and records before publication, and for their solicitor or trade body to check any comparative, superlative or health related claim. It applies the published advertising code and published food information guidance to draft copy. It is not advertising clearance, it is not legal advice, and it is not a finding that the copy complies.

## Built from
- Government Digital Service, "A to Z style guide", GOV.UK publishing guidance, https://guidance.publishing.service.gov.uk/writing-to-gov-uk-standards/style-guides/a-to-z-style-guide/, no publication date shown on the page, read 16 September 2026: sentence case in step 3, and the date range, time range, numeral and telephone number conventions in step 8.
- Home Office, "Readability", Home Office User-Centred Design Manual, https://design.homeoffice.gov.uk/accessibility/written-content/readability, no publication date shown on the page, read 16 September 2026: the maximum reading age of 9 target and the point that simple language is not dumbing down, both in step 4, and the list of readability tools the check in step 4 uses.
- Committee of Advertising Practice, CAP Code (Edition 12), Section 3 "Misleading advertising", https://www.asa.org.uk/type/non_broadcast/code_section/03.html, no publication date shown on the page, read 16 September 2026: rule 3.7 on holding documentary evidence in step 5, and rules 3.1, 3.2, 3.3 and 3.11 in step 6, together with the background statement that the ASA "will take into account the impression created by marketing communications as well as specific claims".
- Food Standards Agency, "Allergen guidance for food businesses", published on GOV.UK 2 April 2020, last updated 17 July 2026, https://www.gov.uk/government/publications/allergen-guidance-for-food-businesses/allergen-guidance-for-food-businesses, read 16 September 2026: the fourteen allergens, the duty to supply allergen information for every item containing one of them, and the signposting route with its "visible, clearly legible and easily accessible" condition, all in step 7.

# site-copy

## You are given
The existing copy of every page of one UK hospitality website, saved as plain text with the URL each file came from. The interview transcript with the owner, head chef or floor team, in their own words, with the speaker and the date. The rewritten copy, page by page, already written by Claude and supplied as finished text. The struck list, one record per removed sentence giving the page, the sentence verbatim, the reason and, where a rule struck it, the CAP Code rule number. The claims register entries, each giving the claim sentence, the page, the claim type and the evidence document supplied with its date, or a note that none was supplied. The allergen wording as finished copy, plus the venue's own allergen matrix where one exists. The style note as finished copy. And the readability results, giving the tool name, the score and the date, per page.

## Produce
Write into a `./site-copy-output/` folder:

1. `pages.csv` with these columns in this order: `page_ref`, `page_url`, `page_purpose`, `words_before`, `words_after`, `first_sentence_before_verbatim`, `first_sentence_after_verbatim`, `answers_page_question_in_first_sentence`, `readability_tool`, `readability_score`, `readability_date`, `sentences_over_25_words`. `page_ref` is P01 upward. `answers_page_question_in_first_sentence` is `yes` or `no` and is copied from the supplied judgement, never decided.
2. `struck.csv` with columns: `struck_ref`, `page_ref`, `sentence_verbatim`, `reason_category`, `reason_supplied`, `cap_rule_cited`, `template_phrase_matched`. `reason_category` is exactly one of `template phrase`, `claim without evidence`, `objective claim capable of substantiation`, `omits material information`, `exaggerates capability`, `scene setting above the answer`, `outdated fact`, `named person or supplier no longer current`. Every row with `reason_category` set to `claim without evidence`, `objective claim capable of substantiation`, `omits material information` or `exaggerates capability` must carry a non-empty `cap_rule_cited`.
3. `template-phrases.csv` with columns: `phrase`, `occurrences_before`, `occurrences_after`, `pages_before`, `pages_after`. One row per phrase in the supplied template list, plus a `TOTAL` row. Any phrase with `occurrences_after` above zero is listed in `gaps.md`.
4. `claims-register.csv` with columns: `claim_ref`, `page_ref`, `claim_sentence_verbatim`, `claim_type`, `evidence_document`, `evidence_date`, `evidence_holder`, `substantiated`, `review_date`. `claim_ref` is C001 upward. `claim_type` is exactly one of `price`, `date or opening time`, `superlative`, `comparative`, `award or accolade`, `provenance or supplier`, `named person`, `quantity`, `health or dietary`, `environmental`, `puffery`. `substantiated` is `yes` or `no`. Dates are DD Month YYYY.
5. `allergen-wording.csv` with columns: `page_ref`, `wording_verbatim`, `route`, `position_on_page`, `behind_collapsed_element`, `telephone_number_adjacent`, `allergen_matrix_supplied`, `matrix_date`. `route` is exactly one of `full allergen information published`, `signposted to staff`, `blanket disclaimer only`, `none`. The three yes or no columns are `yes` or `no`. Any row whose `route` is `blanket disclaimer only` or `none`, or whose `behind_collapsed_element` is `yes`, is listed in `gaps.md`.
6. `style-conformance.csv` with columns: `page_ref`, `convention`, `instances_found`, `instances_conforming`, `non_conforming_examples_verbatim`. `convention` is exactly one of `time range uses to`, `date range uses to`, `sentence case headings`, `telephone number spacing`, `numerals with thousands comma`, `link text describes destination`, `link text not click here`, `link text at end of sentence`.
7. `copy/` holding one file per page containing the supplied rewritten copy reproduced exactly, `style-note.md` containing the supplied style note reproduced exactly, and `interview-transcript.md` containing the supplied transcript reproduced exactly with speaker and date.
8. `gaps.md` is a numbered list of: every page where `answers_page_question_in_first_sentence` is `no`; every template phrase still present after the rewrite with its pages; every claim with `substantiated` set to `no` still appearing in the rewritten copy; every claim with no `evidence_document` or no `evidence_date`; every claim with an `evidence_date` more than twelve months before the build date; every struck row missing a required `cap_rule_cited`; every allergen row flagged above; every style convention with non conforming instances; every page with no readability result; and every named person or supplier appearing in the copy who is not in the claims register with a review date.

## Rules
- Codex measures, records and checks. It never rewrites a sentence, never improves the supplied copy, never shortens a page, never decides that a claim is substantiated and never assigns a CAP Code rule. All of that arrives supplied.
- Never reword, correct, tidy or normalise any quoted sentence, in either the before or the after copy. Spelling, punctuation, capitalisation and any typographic characters are reproduced exactly as supplied, and a typo is recorded in `gaps.md` rather than fixed.
- Never mark a claim substantiated without an evidence document named in the row with its date and its holder. A claim whose evidence is described as available, on file, or with the chef is `no`.
- Never write a benchmark, an industry average, a conversion rate, a bounce rate, a figure for how many customers read a menu page, or any statistic about hospitality websites. No such figure exists for a single independent UK venue and none is to be written into any file.
- Never invent a supplier name, a distance, a farm, an award, a year, a founding date or a member of staff's name, and never carry one forward from the existing copy into the claims register as though it were evidenced. An unevidenced name from the old site is a claim with `substantiated` set to `no`.
- Never record an allergen route as adequate. The route is recorded as one of the four supplied values and the judgement is left to the owner and their environmental health officer.
- Word counts, sentence lengths and occurrence counts are counted, not estimated, and sentence length counts words between full stops including numerals.
- Use British English, GBP and DD Month YYYY dates. No em dashes in any file you write, and any supplied copy containing one is reproduced 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 menu, prices, suppliers and records before publication, and for their solicitor or trade body to check any comparative, superlative or health related claim. It is not advertising clearance, it is not legal advice, and it is not a finding that the copy complies.

## Return
The absolute path of each file written, the row count of each CSV, the word count before and after for every page and the totals, the number of pages whose first sentence does not answer the page question, the number of sentences struck broken down by `reason_category` with the CAP rules cited and their counts, every template phrase still present after the rewrite with the pages it survives on, the number of claims in the register broken down by `claim_type`, the number with `substantiated` set to `no` and every one of those still appearing in the rewritten copy named in full, every claim whose evidence is more than twelve months old, the allergen route recorded for each page, the style conventions with non conforming instances and examples, the readability score per page with the tool named, and the `gaps.md` item count.