← All guides for UK SMEs Practical guide

7 common Claude mistakes UK teams make, and how to fix each one

Chris Duffy

Chris Duffy

Certified Chief AI Officer • • 9 Min Read

The seven Claude mistakes we see most in UK teams are no folder structure, no voice profile, no governance guardrails, a model-training setting nobody has checked, starting every chat from scratch, explaining the same job again instead of building a skill, and taking output on trust. Each one has a fix you can make this week.

All seven come from using Claude like a search box: a question goes in, an answer comes out, and nothing is kept for next time. Each mistake below covers what it looks like, why it costs you, the fix in a few concrete steps, and where we set it up for you.

1. No folder structure

What it looks like. Claude is pointed at the whole shared drive or the Downloads folder. Files are called final v3 (2).docx, there are four price lists, and the client brief is in someone's inbox. Claude picks up the wrong version, or you end up pasting the right one into every chat.

Why it costs you. Claude works from what it can find. Anthropic's Claude 101 course says Claude uses file names to understand and retrieve the right information, so Q4-2024-Brand-Guidelines.pdf helps it more than document1.pdf. A messy folder gives you a polished answer built on last year's prices.

The fix.

  1. Make one working folder for each stream of work, such as Clients/ClientName/, with 01-Brief, 02-Working, 03-Sent and Archive inside it.
  2. Rename the files Claude will use most, date first and then what the file is: 2026-11-Price-List.xlsx, Brand-Voice-Guide.pdf.
  3. When Claude works on files on your computer, give it the folder for the job rather than your whole drive. It asks permission before it changes anything in that folder.
  4. Move superseded files into Archive so Claude stops quoting them.

How we set this up for you: on a Claude training day we set Claude up against the right folders on your team's own machines. For a business of one or two people, the Founders plugin builds the system with you, archive folder included.

2. No voice profile, so it sounds generic

What it looks like. Every draft opens with "I hope this finds you well", uses three adjectives where you would use one, and ends by offering further help. Clients can tell, and so can your team.

Why it costs you. Claude 101 notes that Claude defaults to a helpful, professional tone, which may not match how you write. Generic drafts get rewritten by hand, so the time Claude saved goes straight back into editing.

The fix.

  1. Give Claude your real writing to learn from: connect your email, or put ten sent emails and two documents you are proud of into a folder called Writing-Samples.
  2. Ask for a profile with evidence:Read my sent messages and tell me about my tone and writing style. Quote an example for each habit, and flag anything you're not confident about.
  3. Correct what it got wrong, then ask Claude to save the corrected profile as a skill called /my-voice.
  4. Every time you edit one of its drafts, finish with "add that to /my-voice", so the rules build up week by week.

How we set this up for you: the Founders plugin runs a voice interview as part of its set-up, and on a training day each person builds a skill on their own work.

3. No governance guardrails, so client data sits on personal tools

What it looks like. Three people use Claude on personal accounts, one uses a different chatbot, and nobody can say which client files went where. If there is an AI policy, it was emailed round once.

Why it costs you. A personal account sits outside anything the business can see, set or export. When a client asks where their data went, you have no record to show them, and the rules depend on each person remembering them.

The fix.

  1. Put everyone on one business account, and write down that personal AI accounts are not for work. The AI use policy example in Anthropic's Claude Academy lists this among its hard lines.
  2. Write three short lists: what is encouraged, what needs a conversation first (client names, HR matters, connecting a new system), and what is off-limits.
  3. Add one rule that covers every task: a person checks anything before it goes to a client.
  4. Verify your email domain and turn on Restrict organization creation, so nobody can open a new personal Claude account with a work address.

How we set this up for you: AI Governance writes your AI policy with you and builds the rules into Claude itself, so drafts wait for a person and Claude works only in the folders you approve.

4. The setting that lets your chats train models, left unchecked

What it looks like. Someone opened a personal Claude account, clicked through the privacy choice at sign-up, and has been pasting client emails into it ever since. Nobody has looked at the setting again.

Why it costs you. Anthropic's Privacy Policy says it may use your inputs and outputs to train and improve its models unless you opt out through your account settings. Since its consumer terms update of 28/08/2025, Anthropic trains new models on chats and coding sessions from Free, Pro and Max accounts when that setting is on, and keeps that data for five years instead of 30 days. Team and Enterprise plans sit under Anthropic's Commercial Terms, which that update does not cover, and their conversations are not used for training by default.

The fix.

  1. On any personal account (Free, Pro, Max or Ultra), open your Privacy Settings and check the Model Improvement setting. Make the choice on purpose, and check it again after any change to Anthropic's terms.
  2. Put business use on a Team or Enterprise plan, and move client work off personal accounts.
  3. On a Team plan, an Owner should check Rate chats under Organization settings, Data and privacy. A thumbs up or thumbs down sends that conversation to Anthropic as feedback, so decide whether it is on and say so in your policy.

How we set this up for you: AI Governance settles which account, which settings and which data, and records the decisions in your policy.

5. Starting every chat from scratch

What it looks like. Every conversation starts with the same paragraph about who you are and what the business does. Or nobody writes it, and the answers come back general enough to suit any company.

Why it costs you. Claude works from the conversation, your saved instructions, its memory if that is switched on, and the files it can reach. Anthropic's Cowork course says that outside a project, each session starts fresh apart from your global instructions and what memory has kept. Without saved context, each person rebuilds it by hand, and most stop bothering.

The fix.

  1. Write your standing brief once. Open Settings, search for instructions and choose Instructions for Claude:I'm the operations director of a 40-person logistics firm. "The Monday pack" means the weekly delivery and complaints report. Lead with the answer, keep background to one paragraph, British English.
  2. Make a Project for each stream of work that keeps coming back: a client, the monthly board report, a tender. Give it instructions and the files Claude should always use.
  3. Name the document in your question: "Using 2026-11-Price-List.xlsx, which lines rose more than 5%?"

How we set this up for you: a Claude training day sets Claude up on your team's own machines, and each person spends the afternoon building on their own work.

6. Explaining the same job again instead of building a skill

What it looks like. The weekly update, the proposal, the client follow-up. Each week someone rebuilds the prompt from memory and gets a slightly different result.

Why it costs you. A Claude skill packages your format, your standards and an example, and Claude applies it whenever the task matches. Without one, the quality of the work depends on who wrote the prompt that day.

The fix.

  1. Pick a job you have explained to Claude more than once.
  2. Describe it in one message with three parts, the format, the tone and one real example:Create a skill for my weekly team update. Three sections, wins, blockers and priorities, at most three bullets each. Direct and factual, written for my manager to skim. Here's one that worked well, pasted below.
  3. Test it in a new chat by asking for the task the way you normally would. Open the steps Claude took and check one of them loads your skill. If none does, mention the skill by name or sharpen its description.

Skills need the Create and edit files setting switched on. On Team and Enterprise plans an organisation Owner controls it.

How we set this up for you: every person leaves a training day with a skill for the job they do most, and the best ones are shared so they become everybody's default.

7. Taking a confident answer on trust

What it looks like. A figure goes into a client report because it read well. Nobody traced it back to a source.

Why it costs you. Claude 101 warns that Claude occasionally produces plausible but incorrect information, especially with specific facts or niche topics. A wrong number with your name on it costs more than the hour the draft saved.

The fix.

  1. Ask Claude which documents it drew on, then check the figures against them yourself.
  2. Treat any date, name or quote you cannot trace to something you gave Claude as a flag to check.
  3. Before you hand Claude a recurring task, run it on a past example where you already know the right answer, and compare the two.
  4. Add "flag anything you're not confident about" to every prompt where the answer matters.

How we set this up for you: every training day has a session at 11:30 on spotting the confident wrong answer, before anyone relies on Claude for real work.

Where to start this week

Fix mistakes 3 and 4 first, because they decide which account everything else lives in. Then set up one folder, one standing brief and one skill for the job that eats most of your week. To find that job, use the three signs in the 3-5-4 Method: it repeats several times a week, it follows clear rules, and someone moves information between systems by hand.

A Claude training day is £2,500 for a room of up to 10, plugin included, exc VAT, and works through these fixes on your own work.

Sources

Settings and plan terms change. We checked these sources on 18/11/2026; check the current terms for your plan before you rely on them.

Read next: What a Claude skill is, and how to build one · The Monday Morning Method · Book a call

Common questions

Does Anthropic use my Claude chats to train its models?

On Free, Pro and Max accounts, Anthropic trains new models on chats and coding sessions when the model training setting is on, and its Privacy Policy says it may do so unless you opt out in your account settings. You can change the choice at any time in your Privacy Settings. Team and Enterprise plans sit under Anthropic's Commercial Terms, and their conversations are not used for training by default.

Which Claude mistake should a business fix first?

Fix the account and the guardrails first: one business account, a written rule on which data may go in, and a person checking anything before it reaches a client. The folders, voice profile, Projects and skills all live inside that account, so set them up in the account the business will keep.

Is a personal Claude Pro account fine for client work?

We advise against it. A personal account sits under Anthropic's Consumer Terms, the business cannot see or export what was shared in it, and the model-training choice is made by the individual. Put client work on a Team or Enterprise plan, where an Owner controls the settings and conversations are not used for training by default.

How do I make Claude write in my voice?

Give Claude real writing to learn from, either your sent email through a connector or a folder of samples, and ask it to describe your tone with a quoted example for each habit. Correct what it gets wrong, save the result as a skill such as /my-voice, and add each correction you make to a draft back into that skill.

New guides by email

More in this series: All guides for UK SMEs

Chris Duffy, founder of Ignite AI Solutions

About the author

Chris Duffy Founder and Chief AI Officer, Ignite AI Solutions

Chris is the founder of Ignite AI Solutions, a Certified Chief AI Officer and one of the Top 20 AI Leaders of 2026. He is a regular expert source for Forbes on AI and cybersecurity, and a member of the UKAI Council supporting accredited pathways for upskilling. He was nominated for 3 awards at the National AI Awards 2026, including the Aiconics Award for responsible AI. A UK Special Forces veteran, he now helps UK SMEs adopt AI safely and effectively.

More about Chris · Chris in Forbes · LinkedIn · Substack

Chat on WhatsApp