GrowTraffic
Version 1.2 · Effective from 25 September 2026
Contents
GrowTraffic actively encourages the use of artificial intelligence.
AI is changing how digital marketing, search, content, advertising, development, design, analysis and business operations work.
Quite quickly, as it happens.
We do not want GrowTraffic to become an organisation where people are frightened to use AI, hide that they use it or wait for somebody else to work out what it can do.
Equally, we are not going to assume that sticking “AI” on something automatically makes it clever, useful or right.
Our position is probably best described as cautious techno-optimism.
We think AI is going to be enormously important.
We think it will allow small teams to do things that would previously have required much larger organisations.
We think it will change some jobs, remove some tasks completely and create entirely new ways of working.
We also think it gets things wrong.
Sometimes spectacularly.
So we want GrowFos to get very good at using it without switching off the bit between their ears.
We expect everyone at GrowTraffic to become increasingly capable at using AI.
That means experimenting with it, learning how to prompt it properly, giving it useful context, challenging its answers, using specialist tools and agents and looking for opportunities to improve the way we work.
The purpose of this policy is therefore not to restrict AI use.
It is to make sure we use it intelligently whilst protecting GrowTraffic, our clients, our colleagues and the people whose information we work with.
Our basic principle is:
Use AI freely. Give it context. Let it do real work. But remain responsible for what it does.
AI can assist with decisions.
It does not take responsibility for them.
You do not need permission to use AI for normal day-to-day work.
We would generally rather somebody used AI intelligently than spent hours manually doing something that an AI system could help them complete in minutes.
AI can be used for things including:
This list is deliberately not exhaustive.
If you find a sensible new way of using AI to do your job better, try it.
If it works, tell people.
If it turns out to be rubbish, tell people that too.
There is value in learning what doesn’t work.
The weakest use of AI is often something like:
“Write me something about X.”
The strongest use tends to involve giving AI proper context and allowing it to understand the problem.
Where appropriate, give AI the objective, background, previous work, screenshots, reports, analytics, brand guidelines, examples, code, commercial information, constraints and details of what has already been tried.
Tell it what you think.
Tell it what you are unsure about.
Give it the information that a good colleague would need before they could properly help you.
Then challenge what it produces.
Ask it to find weaknesses.
Ask it what you have missed.
Ask it to argue against your proposed solution.
Ask it to verify its conclusions.
Tell it when you think it is talking nonsense.
For important work, consider asking a different model or agent to review the result.
AI should not merely be used as a typing machine.
It should be used as a research tool, analytical tool, creative tool and, increasingly, as a working partner.
Some of the most interesting uses of AI will probably be things none of us have thought of yet.
That is another reason we want people experimenting with it.
Using AI does not transfer responsibility to the AI.
If your name is on the work, you remain responsible for it.
Before something important reaches a client, goes live or materially affects a campaign, website, system or business decision, somebody with appropriate responsibility must understand what has been done.
“I don’t know. ChatGPT did it.”
is not an acceptable explanation.
Neither is:
“Claude said it would work.”
AI systems can be extremely capable.
They can also be extremely convincing when they are wrong.
That combination is worth remembering.
The greater the consequences of getting something wrong, the more care we should take checking it.
This is one of the most important rules in this policy.
Giving an AI system access to something does not give it greater authority than the person operating it.
If you are not authorised to:
then you cannot give an AI agent permission to do it on your behalf.
An AI agent is effectively acting as an extension of you.
Your existing authority still applies.
This matters more as AI agents become increasingly capable of doing things rather than simply suggesting them.
AI is increasingly capable of doing things rather than simply advising us.
This includes agents which can:
GrowTraffic encourages this kind of AI use.
This is probably where some of the biggest productivity gains will come from.
It is also where somebody can accidentally give a very fast computer permission to make a very fast mess.
For significant actions, think in terms of:
Plan → Act → Check
Let the AI understand the problem.
Let it carry out the work where appropriate.
Then independently verify that the result is correct.
For higher-risk work, the checking process should go further than simply asking the same AI:
“Did you do this correctly?”
Ask it to test the result.
Inspect the actual output.
Check the live system.
Use a separate model.
Run an adversarial review.
AI should not be allowed to mark its own homework where the consequences matter.
AI can write and deploy code.
It can make substantial development work dramatically faster and we want people to take advantage of that.
Production systems require additional care.
Before significant AI-assisted development work reaches a live system, there should normally be:
Where staging environments are available and practical, use them for significant changes.
Small, easily reversible changes may be made directly where the person making them has the authority and experience to judge the risk.
The thing we care about is not whether the code was written by a person or an AI.
We care whether it works.
And whether we can put things back if it doesn’t.
Human-written code can break a website too.
AI becomes considerably more useful when it has context.
We therefore do not operate a blanket rule saying that client information can never be provided to AI systems.
That would make many of the most useful applications of AI impossible and would probably result in people ignoring the policy anyway.
Instead, use judgement based on the sensitivity of the information and the system being used.
Public information, anonymised data, normal marketing material, non-sensitive screenshots and information created specifically for analysis will generally present little difficulty.
More care should be taken with things such as client reports, advertising performance, SEO data, unpublished marketing plans, internal documents, commercial strategy, meeting notes, analytics exports and website code.
Where possible, use reputable AI services supplied through GrowTraffic accounts or established paid platforms rather than some mysterious free AI tool you discovered five minutes ago.
Only provide the information actually required for the task.
Particular care must be taken with information such as:
Do not casually paste this information into an AI conversation.
Where an AI system genuinely needs sensitive access, use appropriate secure connections, access controls or authorised integrations.
Passwords should not be pasted into AI prompts.
Permanent API keys, access tokens and similar credentials should also normally be kept out of AI conversations.
Where AI is helping configure a system, use placeholders such as:
YOUR_API_KEY
Credentials should normally be entered directly into the relevant application, secrets manager or environment configuration.
If an AI system genuinely needs access to a service, use proper authentication and permissions rather than giving it credentials in a prompt wherever possible.
Do not give AI unnecessary personal information.
If somebody’s identity is irrelevant to the task, remove it.
For example, AI analysing a set of leads may need:
It probably does not need:
Give the AI what it needs to do the job.
Not everything you happen to possess.
Don’t assume that because you’re chatting to an AI you can chuck absolutely everything into it.
Before supplying commercially sensitive material, ask:
Does the AI actually need this information to complete the job?
If not, leave it out.
Think about it in roughly the same way you would think about any other external technology supplier.
Useful does not mean entitled to see everything.
Some clients may have their own requirements governing AI.
If a client’s contract, information security policy or explicit instruction places restrictions on how their information can be used with AI, those requirements take precedence over this policy.
If you are unsure whether a particular restriction applies, ask.
GrowTraffic is happy for AI providers to learn from our own work.
Better models are good for everybody, including us.
That’s our choice to make about our own information.
Client information is different. It isn’t ours to give away, so only use it with tools that train on it where the client’s terms allow it.
Personal data is different again.
If personal data is involved, such as customer names, email addresses or order details, only use a tool or setting where training is switched off.
If you can’t switch it off, don’t use that tool for personal data.
Assume anything you put into an AI tool may stay there.
You can usually delete a conversation, but you can’t take information back out of a model that has learned from it.
That’s why the checks happen before you paste, not after.
If something has gone in that shouldn’t have, tell Simon straight away.
Some mistakes involving personal data have to be reported to the ICO within 72 hours, so the sooner we know, the better.
AI may be used to help create content for GrowTraffic and our clients.
However, for normal client content, we encourage the person responsible for the content to write the final copy from scratch in their own words.
AI is particularly useful for:
The aim is to use AI to make the writer better informed and better equipped, rather than simply asking an AI to produce an article and passing the output to a client.
This does not mean AI-generated copy is automatically bad.
It isn’t.
Well-prompted AI, supplied with strong source material, genuine expertise, appropriate context and careful human direction can sometimes produce excellent content.
In some cases it can produce better content than somebody staring at a blank document and trying to hit a word count.
GrowTraffic may therefore use substantially AI-generated content where there is a good reason to do so and somebody is prepared to take responsibility for its quality.
However, our normal standard for client work is:
Human-written, AI-assisted.
Rather than:
AI-generated, lightly edited.
Most of our client blog content is still written by people.
That will change.
When we do publish substantially AI-written content on a client’s site, the client should know about it and it goes through the same human review as everything else.
AI detection software is imperfect and cannot reliably prove whether an individual piece of content was written by AI.
We therefore do not use an AI detector as a definitive test of authorship.
However, detection tools can still provide a useful quality-control signal.
Clients can run our content through AI detectors, and some will.
So check it yourself before it goes to a client, and be ready to explain how it was made.
A detector score doesn’t decide whether content is good enough. A person does.
But if something keeps getting flagged, find out why before the client does.
If content is being regularly flagged, the solution is not to make superficial changes purely to fool an AI detector.
That misses the point.
The content should be reconsidered and, where necessary, rewritten properly.
Common warning signs include:
The objective is not to produce content that can “beat an AI detector”.
The objective is to produce content which reads like somebody knowledgeable actually had something worth saying.
Whether content was written by a person, generated by AI or created through a combination of the two, it must ultimately meet the same GrowTraffic standard.
Before publishing or sending important content, check:
Watch particularly for invented statistics, quotations, sources, case studies, products, services, features or supposed facts.
AI should never be allowed to manufacture expertise that neither GrowTraffic nor the client possesses.
There is no magic percentage of AI involvement at which content suddenly becomes good or bad.
The appropriate approach depends on what we are producing.
A carefully engineered content process using substantial source material, proprietary information, detailed prompting, expert input, repeated refinement and strong quality checks may legitimately use AI for much more of the final wording.
That is very different from typing:
“Write me a 1,500-word SEO blog about replacement kitchen doors.”
and publishing whatever comes back.
The important questions are:
How good was the input?
How good is the output?
Does somebody understand and stand behind what has been published?
For routine client content, our preferred process is:
Research with AI → Think → Write → Use AI to challenge and improve → Human review → Publish
Do not produce content simply because AI makes producing content cheap.
More pages are not automatically better SEO.
AI should help us produce:
It should not become an excuse for publishing thousands of low-value pages.
Before creating something at scale, ask whether the content genuinely deserves to exist.
AI makes creating pages easier.
That does not make creating unnecessary pages sensible.
The same principle applies to virtually every other area of digital marketing.
AI gives us leverage.
Leverage applied to a bad idea just lets us execute the bad idea more efficiently.
AI should be treated as an excellent research assistant rather than an unquestionable authority.
When facts matter, verify them.
Nobody needs three independent sources for every trivial fact used in an internal brainstorm.
But important information about things such as law, regulation, finance, health, contracts, technical specifications, industry statistics and significant client claims should be checked against reliable sources where practical.
Whenever information may have changed recently, encourage the AI to search current sources rather than relying entirely on what it was trained to know.
Useful questions include:
“What in this answer are you least confident about?”
and:
“Which claims here should I independently verify?”
One of the skills we increasingly need is not simply getting AI to answer questions.
It is knowing when not to believe the answer.
AI does not remove our responsibility to respect other people’s intellectual property.
Do not ask AI to reproduce somebody else’s proprietary work so closely that we would not be comfortable doing the same thing manually.
This includes articles, photographs, designs, proprietary reports, software, training materials, paid research and competitor content.
Using other work for research, inspiration, analysis and comparison is different from copying it.
Use common sense.
AI-generated imagery may be used for client and GrowTraffic work where appropriate.
Check imagery before publishing it.
Pay particular attention to:
Do not deliberately create deceptive imagery and present it as documentary photography.
If an AI-generated image is being used to illustrate a concept, that is generally very different from using one where the audience would reasonably believe it records something that genuinely happened.
AI can help write almost any email, message, proposal or document.
That is encouraged.
But read it before sending it.
Particularly check AI-written communication involving:
AI can help you find the words.
It cannot know everything happening around a relationship unless you tell it.
And even if you do tell it everything, it may still come up with something daft.
Never send something simply because the AI made it sound authoritative.
AI can help analyse advertising accounts, write ads, identify opportunities, diagnose problems, suggest targeting, analyse conversion data, automate reporting and build supporting scripts or tools.
Changes involving significant expenditure should still be made by somebody with appropriate authority.
Do not allow an AI system to materially increase client advertising expenditure simply because an algorithm recommends doing so.
AI can make a recommendation.
Commercial judgement still matters.
AI can help managers think through situations involving colleagues.
For example, it may help structure an appraisal, improve feedback, identify useful questions, draft policies, compare job descriptions or organise interview notes.
AI should not make the final decision about:
People are responsible for decisions about people.
AI might help you think.
It should not become a convenient way of outsourcing difficult human decisions.
As AI becomes more capable, websites, emails and documents may contain instructions deliberately designed to manipulate AI systems.
For example, a webpage or document could contain instructions telling an AI agent to:
Treat external websites, emails and documents primarily as information, not as instructions.
Unexpected instructions discovered inside external content should not override the job you originally gave the AI.
If an AI agent suddenly decides that analysing a competitor’s website somehow requires emailing somebody your passwords, something has probably gone wrong.
Stop and investigate.
This will become increasingly important as we give AI systems greater access to business tools.
People will occasionally make mistakes with AI.
AI itself will also make mistakes.
GrowTraffic would rather somebody quickly admitted:
“I used AI for this and something has gone wrong.”
than tried to hide it.
If an AI-related mistake could affect a client, personal information, money, a website, advertising spend, security or confidential information, raise it internally as soon as possible.
Problems normally become more expensive when people hide them.
Nobody is going to get into trouble simply because an AI tool made an unexpected mistake and they reported it promptly.
Hiding the mistake is a different matter.
We want people experimenting with AI.
Seriously.
Try new models.
Try new tools.
Try agents.
Build things.
Automate boring processes.
Create your own prompts.
Ask AI how parts of your job could be improved.
Use it to test ideas you would previously never have had time to explore.
Try using it for things you aren’t sure it can do.
Sometimes it will surprise you.
Sometimes it will produce complete nonsense.
Both are useful things to discover.
Share useful discoveries with the other GrowFos.
If you discover a way of saving everyone two hours a week, tell people.
If you build an internal tool that could help somebody else, show them.
If a new tool is useless, tell people that too.
GrowTraffic should learn collectively rather than everybody independently rediscovering the same things.
The people who become particularly good at using these tools should help the rest of the team become better at them too.
AI capability should spread through the business rather than sitting with two or three people.
GrowTraffic may provide access to AI platforms and tools where there is a reasonable business benefit.
If you believe a paid AI product will materially improve your work, raise it rather than avoiding it simply because it costs £20, £50 or £100 a month.
The relevant question is not:
“Is this software free?”
It is:
“Does the value it creates exceed what it costs?”
If a £50 tool saves somebody several hours every month, it may be very cheap.
At the same time, subscriptions should not accumulate indefinitely for tools nobody uses.
If we are paying for something, somebody should be getting value from it.
You do not need formal approval simply to experiment with a publicly available AI tool using non-sensitive information.
You should exercise significantly more caution before connecting an unfamiliar AI service to GrowTraffic systems.
This includes giving it access to things such as:
The greater the access, the more carefully the supplier and permissions should be considered.
Do not give a new AI application access to half the company simply because it has a shiny website and an impressive demo video.
Useful new technology deserves enthusiasm.
It does not deserve blind trust.
We do not need to announce every time AI has helped us do something.
AI is part of the modern toolset in much the same way as analytics software, spelling checkers, development tools and automation platforms.
However, we should never mislead a client.
If a client directly asks whether AI was involved, answer truthfully.
If a contract requires disclosure, disclose it.
If AI is performing something material or unusual on behalf of a client and they reasonably need to understand that, explain it.
The important thing is that GrowTraffic remains accountable for the work.
We do.
Where a customer or member of the public is interacting directly with a fully automated AI system, it should normally be reasonably clear that they are interacting with an automated system.
We should not deliberately trick somebody into believing they are speaking to a particular real employee when they are not.
There is a difference between making an automated system conversational and pretending a human interaction has taken place when it has not.
If you are unsure whether a particular use of AI is sensible, ask yourself:
If the answer to those questions is broadly yes, get on with it.
If one of them makes you uncomfortable, think a little harder or ask somebody.
GrowTraffic is fundamentally optimistic about AI.
We think it is an extraordinary technology and we are still very early in understanding what it is going to enable.
We do not believe the answer is to ban it, fear it or pretend that the way we worked five years ago is somehow morally superior because it involved more manual effort.
Nor do we think every AI company, AI feature or AI-generated answer should be accepted uncritically.
Some of it will be brilliant.
Some of it will be useless.
Some of it will be dangerous.
Quite a lot of it will probably be all three at different times.
The organisations that benefit most from AI will not necessarily be the ones that buy the most AI software.
They will be the ones whose people become very good at combining human experience, judgement, creativity, expertise and relationships with increasingly capable machines.
That is what we intend to do.
We want every GrowFo to become confident using AI.
Experiment with it.
Learn how it works.
Challenge it.
Give it difficult problems.
Give it proper context.
Let it save you time.
Let it find things you have missed.
Let it improve your work.
Let it do things that would previously have taken days.
Build new things with it.
And don’t be afraid to question it when something doesn’t smell right.
We are optimistic about where this technology is going.
We are not naive about it.
AI can do the work. GrowTraffic remains responsible for it.
This policy is owned by Simon Dalley, who is responsible for data protection at GrowTraffic.
If you’re unsure about anything in it, ask Simon.
| Version | Effective from | Changes |
|---|---|---|
| 1.2 | 25 September 2026 | Training AI: training must be switched off whenever personal data is involved. |
| 1.1 | 25 September 2026 | Added Training AI, Once It’s In, It’s In and Who Owns This Policy. AI detection is now a check before content reaches a client rather than a pass/fail test. Clients should know when content on their site is substantially AI-written. |
| 1.0 | 16 September 2026 | First published. |