Claude marketing and Claude at work
Anthropic ruled ads out of Claude in February 2026 and has held that line since. No sponsored slot, no bidding, nothing an agency can switch on for you. What is left is being described accurately enough to be worth citing, and having Claude actually do work in your business while your competitors are still pasting things into an empty box.
You are probably using it for emails, captions and the odd awkward reply. That is a fraction of what you are already paying for. The gap between a tool you chat to and a tool that does Thursday afternoon for you is not the model. It is the brief, and the brief is learnable.
Both run in your browser. Nothing you type is stored or sent to us, and no email address is needed to see your result.
Take last quarter's enquiry spreadsheet, work out which sources are actually converting, and write it up for Monday's meeting.
What we work on around Claude
Two halves of one job: what Claude says about you, and what Claude does for you.
Where we come in
Nobody has failed at anything. The licences are paid, a few people are keen, and yet nothing about the week has changed.
You are not doing it wrong. You are doing the easy tenth of it, which is what everybody does at first.
The keen ones went first and got value in a fortnight. Everybody else opened it once, saw a blank page and went back to doing the job the way they already knew how. That is not resistance. Nobody told them which of their tasks were on the table, and a blank page is a terrible brief.
Because it is working from nothing. No price list, no past proposals, no tone of voice, no idea who your customer is or what you refuse to say. Give it your material and the output stops being generic within a day. This is the most common complaint about AI and the easiest one to fix.
Client data, contracts, staff records, the pricing model. Somebody has already pasted something they should not have, and somebody else is refusing to use it at all because there is no rule either way. One page of policy fixes both problems in an afternoon.
Quick questions, quick answers, a bit of tidying up. Useful, and nowhere near the point. The jobs worth handing over are the ones you dread: the quote you keep putting off, the invoice chase, the four hours of admin that eat your Friday.
Same explanation of your business, same tone instructions, same corrections, typed again on Tuesday. Everything you keep repeating can be stored once and reused. Ten minutes of setting up saves the same ten minutes every week for the rest of the year.
Assistants describe businesses using whatever independent material exists about them. If the only place you are described is your own website, there is nothing to corroborate and you get left out of the shortlist. Fixable, and it does not need you to post daily.
First, a definition
They get talked about as one subject and they are not. Different budgets, different people, different measures of whether it worked. Sorting them out is most of what a first call is for.
The unpaid one, and the only marketing lever that exists here. When Claude searches the web it cites what it reads, so the question is whether anything it can read describes you accurately and specifically enough to be worth quoting.
The one that actually pays, and the reason most of our Claude work exists. Not chatting. Finishing things: the report, the spreadsheet, the deck, the ninety enquiry replies, the research nobody has time for.
Where Claude stops being an assistant and becomes part of what you sell: a tool on your site, an agent that handles a queue, a process that runs without anybody opening a browser at all.
We do all three. Most first conversations end with us saying start with the second, because it pays for the other two and you will feel it inside a fortnight.
The brief lab
Most disappointing output is a brief problem rather than a model problem. Put a real instruction in, add the line you use to describe your business, and pick what you want out of the other end. The panel shows what a thin brief forces Claude to invent, what the same request looks like once it has something to work with, and the rebuilt brief itself, which is yours to take away.
Do not polish it for us. The messy version is the useful one, because that is the one you type at half past four on a Thursday.
What you want out
Everything here runs in your browser. Nothing is stored, nothing is sent to us, and nothing is sent to Claude either.
Write something about our new opening hours
Switch between the tabs. The gap between the first two is the entire service.
The Claude readiness scorecard
Four pillars, weighted evenly. Foundations first, because nothing above it can be judged until the rules and the accounts are settled. Answer honestly rather than generously: a low score here is cheap to move, which is rather the point.
What we actually do
Delivered in that order, because each one makes the next cheaper. We are hands on with all of it and we hand it over rather than sitting on it.
We sit with you for an hour and list what your week is actually made of. Then we mark the jobs that repeat, the jobs that need judgement and the jobs that are just typing. Only the third group moves first, and there is always more of it than people expect.
Your prices, proposals, tone of voice, the objections you hear weekly and the answers that work, loaded once into a Project so every conversation starts from your business rather than from nothing. This is the step that stops output sounding like everyone else's.
A good brief written once, saved, and reusable by anybody in the team. The monthly report, the quote, the listing, the follow-up sequence. Your best process, available to your newest employee on their first morning.
Mail, calendar, drive, Slack and the tools you already pay for, connected properly and scoped tightly. The difference between asking Claude to write something and asking it to go and find the thing first.
The marketing half. Making sure the facts about you exist somewhere independent, in plain language, in a form a model can read and repeat without hedging. Slow, cumulative, and it keeps working after you stop paying for it.
One page on what never goes in, an hour with the team on the three habits that matter, and a written record of how everything was set up. Then we get out of the way. If you need us monthly, something has been built wrong.
The part that changes the strategy
Every other page on this site has a paid lever somewhere. This one does not, and that is worth understanding properly before you spend anything.
Anthropic published its position in February 2026 and put it on television at the Super Bowl. Claude stays ad-free.
The reasoning was that advertising inside an assistant creates an incentive to keep you talking rather than to be useful, and that once those incentives arrive they tend to spread. Whatever you make of the motive, the practical consequence for you is simple. There is no sponsored slot under a Claude answer, no auction, no budget that shortcuts the work, and no agency on earth that can put you in front of somebody who did not ask about you.
Which leaves two honest levers, and they are the two this page is about. Be described accurately enough in enough independent places that you are worth naming when somebody asks. And use the thing properly, so that the hours it gives back are real rather than theoretical. Nobody can sell you a shortcut past either.
There is a second consequence people miss. Because Claude is paid for by subscriptions rather than attention, the people using it are mostly at work, on a paid plan, in the middle of a task. That is a small and unusually valuable audience, and it behaves nothing like a free chatbot audience.
Checked September 2026. Anthropic moves quickly and plan details change, so treat anything specific here as a snapshot and confirm it before you build a budget on it.
The arithmetic
Five sliders and straight arithmetic. No forecasting, no case studies, no percentage lifted from a vendor's blog. The number that matters is not what gets drafted faster, it is what is left after somebody has checked the work.
The review slider is the one that tells the truth. Every serious estimate of this work has the same shape: drafting collapses, checking does not, and the teams that pretend otherwise ship something wrong within a month and then swear off the whole idea. Thirty per cent is a fair starting assumption for work that goes to a customer. For anything internal you can go lower. For anything regulated, go higher and be glad you did.
The other honest note is that hours saved are not money until somebody decides what they are for. If the answer is that everybody gets an easier week, that is a perfectly good answer, but write it down as the goal rather than discovering it by accident six months later.
Licence cost assumes roughly £18 a person a month. Everything else comes from your own sliders. Nothing here is a forecast.
How it runs
Nothing here needs a transformation programme, a steering group or a new job title. It needs one person inside the business who cares, and about two hours a week for the first month.
The week audit, the rule about what goes in, and the right plan for what you actually do. We pick three repeating jobs, not thirty, and we pick the ones somebody already resents doing.
Your material goes in. Prices, proposals, tone, the questions you answer weekly. The three jobs get proper briefs, get tested against work you have already done, and get corrected until the output is genuinely usable.
Connectors where they earn their place, and the first job that runs end to end without anybody copying anything between windows. This is usually the week somebody in the team stops being sceptical.
The marketing half: the sources that describe you, the pages written as answers, the measurement. Then handover, documented, so you can add the next five jobs yourself without ringing us.
One session and a written plan. For businesses that want to know what is worth doing before committing to anybody doing it.
The ninety days above, delivered. We build it, your team learns it by using it, and it belongs to you at the end.
For businesses already running well who want somebody to ask when something new lands, which at the moment is roughly monthly.
Before you ask
Mostly a different thing, and you do not have to choose. ChatGPT is where a growing number of your customers ask questions, so it matters commercially whether it names you. Claude is where a growing number of businesses do the work itself, and it is unusually good at long documents, spreadsheets, code and anything that needs care rather than speed. Plenty of our clients run both and use each for what it is best at. If you want the other half of the argument, the ChatGPT page covers it properly.
Nobody can guarantee that, and anybody promising it is selling something. What can be done is removing the reasons a business gets left out: facts that exist nowhere except your own site, claims no independent source supports, blocked crawlers, and no page that answers the buyer's question in the buyer's own words. Do that and you are in the pool of businesses that can be named. It is the same honest position we take on every search surface.
On Team and Enterprise agreements, your content is excluded from training at the contract level, which is one of the reasons those plans exist. On individual plans the settings matter and are worth checking rather than assuming. Separately, the rule we write for every client is simpler than the licence terms: decide what categories of information are never pasted in anywhere, write it on one page, and tell everybody. Most real incidents are not about model training. They are about somebody sharing a chat that had a client's figures in it.
It can, and pretending otherwise would be silly. The risk drops sharply when it is working from your material rather than from memory, when it has been asked to cite what it used, and when a human checks anything that leaves the building. That is why the review slider in the calculator above exists and why we never build a process where nobody reads the output. Use it for drafting, structuring, summarising and analysis of things you can verify. Do not use it as an oracle.
Often more than for a large company, because there is nobody else to delegate to and no committee to get through. One person with three good Skills and their prices loaded into a Project gets more back per pound than a hundred-seat rollout that nobody was trained on. Start with the job you dread most. That is nearly always the right first candidate.
You do not need to be. The lab above is the whole skill in miniature: say who it is for, give it the context it cannot know, say what shape you want back, point at something to work from, and say what to avoid. Everything beyond that is us doing the fiddly parts once so your team never has to think about them.
Theirs is a subscription, roughly the price of a paid seat per person per month, with heavier plans for people running it hard all day. Ours depends on which of the three engagements above fits, and we will tell you on the first call rather than after a discovery phase. If the honest answer is that you do not need us and should spend two afternoons doing it yourself, we will say that too. It happens more often than you would expect.
No. The brief lab, the scorecard and the calculator all run in your browser. Nothing is sent to us, nothing is sent to Anthropic, and it disappears when you close the tab. If you copy your brief or your score, that copy goes to your own clipboard and nowhere else. Our privacy notice covers the rest of the site.
Claude and Anthropic are trademarks of Anthropic PBC. TWV Media is an independent agency and is not affiliated with, endorsed by or sponsored by Anthropic. Product details on this page were checked in September 2026 and change frequently.
Also worth a look
Book a call
Bring one job you hate doing. We will brief it live on the call, show you what comes back, and tell you whether it is worth handing over, whether or not you ever hire us.
Book a free 20-minute call