AI Services
Generative AI content with human quality control
Generative AI cuts content production time dramatically, but raw AI output is not publishable. We use AI for speed and volume, then apply human editing, fact-checking and brand review to everything, so nothing generic or wrong ships under your name.
Who it is for: Marketing teams that need more content than their budget used to buy, without the generic tone and errors that give raw AI output away.
Everything in this service
- AI-assisted copy edited by a human, checked against your voice
- Product and campaign imagery generated and retouched
- AI-assisted video for ads and social formats
- Fact-checking on every claim before publication
- A documented workflow you can audit
What to expect
- Content volume up without headcount up
- A consistent voice across everything published
- Zero unchecked AI claims going out under your brand
How generative ai content actually works
Human review is a requirement, not an option
Every piece that leaves our workflow has been read, changed and approved by a person. That is not a service level we offer at a higher tier. It is the only way we work, because the failure modes of unreviewed generation are exactly the ones that damage a brand: a confident factual error, a claim you cannot support, a phrase that belongs to a competitor, an image with a detail that is wrong in a way the model cannot see.
The review is substantive rather than a glance. An editor checks the facts, cuts the padding, rewrites the parts that sound like nobody, verifies that any claim about your product matches what your product actually does, and confirms it reads as your business rather than as generic industry writing. Pieces that need too much of that are discarded and briefed again, because rewriting a bad draft line by line takes longer than starting from a better one.
This has a cost, and it is the honest reason AI does not make content free. It makes drafting and variation fast, which is a real saving. Editing, verifying and judging remain human work at roughly human speed. Anyone selling volume without that step is selling you the risk, not the saving.
- Every piece is read, edited and approved by a person
- Editors verify claims against what your product actually does
- Pieces needing heavy rescue are rebriefed, not patched
- Generation saves drafting time, not verification time
- Volume without review transfers risk to your brand
We will not produce scaled low-quality content
Publishing large volumes of thin, generated pages is a documented cause of traffic loss. Search engines have acted specifically against mass-produced content made primarily to rank rather than to help, and the outcome for sites that did it has ranged from pages being ignored to sitewide damage. We will not build that for a client, and we will say so plainly if it is what is being asked for.
The distinction that matters is purpose and quality rather than tooling. Google has been consistent that it rewards helpful content however it was produced, and acts against filler however it was produced. A page written with AI assistance that answers a real question accurately, from someone with genuine knowledge of the subject, is fine. Two hundred near-identical pages spun to target keyword variants are not, and the fact that a human clicked publish does not change that.
What this means in practice is fewer, better pages, and honesty about the ceiling. Generation helps you produce a strong draft faster, explore more angles, and handle the mechanical parts of production. It does not manufacture the expertise, the original observation or the specific detail that makes a page worth reading, and it cannot substitute for having something to say.
- Mass-produced filler has cost sites traffic, we will not build it
- Quality and purpose decide the outcome, not the tool used
- Near-duplicate pages targeting keyword variants are the classic failure
- Fewer, better pages beat volume for its own sake
- Generation speeds production, it does not supply expertise
Where generation helps and where it damages quality
Generation earns its place in specific parts of the process. Producing multiple angles on a brief quickly, so the best one can be chosen rather than the first one. Drafting the structural, mechanical sections of a longer piece. Adapting one piece of content into formats for different channels. Producing variations of ad copy or subject lines for testing. Generating background imagery, textures, mockups and concept visuals. Cleaning up transcripts. Summarising research so an expert can work from it.
It damages quality in equally specific places. Anything requiring first-hand experience, because the model does not have any and will produce a fluent imitation of it. Technical explanation where precision matters, since plausible and correct are close enough to be hard to spot. Anything with numbers, dates, prices or specifications. Opinion pieces, which read as opinion-shaped rather than as an opinion. Images of real people or places connected to your business. And anything where being the same as everyone else is the actual risk, because the model is trained on everyone else.
The rule we work to is simple. Generation handles production. People supply the substance: the experience, the judgment, the specifics, the point of view. When a brief has no substance behind it, the honest answer is that the piece should not be made rather than that AI should make it.
- Good for variations, structure, adaptation and concept visuals
- Poor for experience, precision, numbers and genuine opinion
- Never generate images of real people or your real premises
- The model is trained on everyone else, so sameness is the risk
- If there is no substance behind the brief, do not publish it
Fact-checking as a defined step
Generated text states things confidently whether or not they are true, and the errors are designed to look ordinary. That makes checking a scheduled step with a method rather than a habit of reading carefully.
The method is unglamorous. Every factual claim is identified and traced to a source we have actually opened. Numbers, dates and prices are checked against a primary source rather than an article quoting one. Statistics without a traceable origin are cut rather than softened, because a hedged version of an invented number is still an invented number. Claims about your own business are verified against your material, not assumed. Quotes are checked to a real source or removed. Names and titles are confirmed. And anything about a regulated area is either removed or referred to someone qualified.
The same applies to our own marketing about AI, which is a field full of numbers that trace only to other blogs. We would rather write the direction than repeat a figure we cannot stand behind, and we apply that standard to your content too. It occasionally makes a piece less punchy. It also means nothing we publish under your name is waiting to be found out.
- Trace every claim to a source someone has actually read
- Check numbers against primary sources, not articles quoting them
- Cut untraceable statistics rather than hedging them
- Verify claims about your own products against your own material
- Refer regulated subjects to someone qualified
Controlling brand voice across generated work
The tell of unedited AI writing is not a single phrase. It is a texture: sentences of similar length, a habit of announcing what the paragraph will do, symmetrical structures, and a general reluctance to say anything specific. Readers recognise it even when they cannot name it, and it reads as a business with nothing particular to say.
Controlling for it needs a written voice standard with real examples, not adjectives. Sentence rhythm, how formal you are, whether you address the reader directly, the words you use for your own products, and a banned list of the filler that creeps into everything. Examples of your own writing at its best are worth more than any description. That standard guides both the brief and the edit, and it is what an editor grades against.
Even with a good standard, the last mile is human. Editors cut the throat-clearing, break the rhythm, replace generic statements with specifics from your business, and remove the words on the banned list. The result should be indistinguishable from your team writing well, because that is the only version worth publishing.
- Write a voice standard with real examples, not adjectives
- Keep a banned-words list and enforce it in editing
- Break uniform sentence rhythm, it is the clearest tell
- Replace generic statements with specifics only you can supply
- Grade every piece against the standard before it ships
Disclosure, testimonials and the rules that apply
Some uses of generated content are not a quality question but a legal one. Reviews and testimonials are the clearest case. Fabricated reviews, testimonials attributed to people who did not give them, and generated endorsements presented as genuine customer feedback carry real regulatory exposure, and the rules in this area have teeth. We do not produce them, and we would advise any client asking for them to take it to a qualified adviser first, because the answer is not going to be one we can improve on.
The same applies to generated images of people used as if they were customers or staff, generated case studies presented as real projects, and generated results attributed to work that did not happen. Placeholder proof is fine while a site is being built. Publishing it as real is not.
On disclosure of AI assistance more generally, the position we take is practical. We tell you exactly where AI is used in producing your content, so you can make your own disclosure decisions with full information. Where content appears under a named author, that person should have genuinely reviewed and stood behind it. Where an image is generated rather than photographed, that is worth being straightforward about if a reader could reasonably assume otherwise. The cost of disclosure is small. The cost of being caught concealing it is not.
- We do not generate reviews, testimonials or endorsements
- Never present generated people or projects as real customers or work
- Take questions about review rules to a qualified adviser
- A named author must genuinely stand behind the piece
- Be straightforward when an image is generated rather than photographed
The workflow, and what you should be able to audit
A defensible workflow can be described in a paragraph and checked by someone outside it. Ours runs in the same order every time. The standard is set first: voice, visual style, claims that are off limits, subjects that require a specialist. A brief is written per piece with the angle, audience and the specific points that must appear. Generation produces drafts and options against that brief. An editor rewrites and grades. Facts are checked and sourced. A brand review confirms it sounds like you. Then it publishes, and performance feeds back into the next brief.
You should be able to see any of that. Which pieces were AI-assisted and in what way, who edited each one, what sources back the claims, and what was rejected. Keeping that record costs almost nothing during production and is extremely useful when someone asks a hard question about a published claim six months later.
It also makes improvement possible. Patterns in what editors keep rejecting tell you the briefs are wrong. Patterns in what performs tell you the angles are right. Over time the briefs get sharper and the amount of editing per piece falls, which is where the actual efficiency of this approach comes from. It is not from removing the human. It is from giving the human less to fix.
- Standard, brief, generate, edit, verify, brand review, publish
- Record who edited each piece and what sources back the claims
- Keep rejected work, the patterns tell you the briefs need fixing
- Feed performance back into briefs rather than into templates
- Efficiency comes from better briefs, not from skipping review
A clear path, step by step
- 01
Set the standard
Your voice, visual style and non-negotiables documented before anything is generated.
- 02
Generate
AI produces volume and variations fast, guided by briefs, not left to wander.
- 03
Edit and verify
Humans rewrite, retouch and fact-check. Anything below standard is redone or discarded.
- 04
Publish and learn
Performance feeds back into the briefs, so quality and speed both improve.
Why choose us for this
Human review on every piece, no exceptions
We disclose exactly how AI is used in your workflow
Editors trained on your voice, not a generic style
Common questions
Will people be able to tell the content is AI-assisted?
Not if the process is right. What gives AI away is the unedited output: generic phrasing, confident errors, uniform rhythm. Human editing against a real voice guide removes exactly that, which is why it is mandatory in our workflow.
Does Google penalise AI content?
Google says it rewards helpful content and penalises mass-produced filler, however it was made. AI-assisted content that is edited, accurate and genuinely useful ranks. Unedited AI spam is what gets hit.
Who owns the content you produce?
You do. We work with tools whose terms support commercial use and hand over full rights to the finished, human-edited work.
Explore related work
Want this for your business?
Book a free visibility call and I will tell you honestly whether I can help.
How this is delivered
One person leads every project. Where a job genuinely needs a specialist, I bring in people I have worked with before and manage them, so you get one point of contact and one invoice rather than three suppliers blaming each other.
- You talk to the person responsible for the work, not an account manager
- Specialists are briefed and managed by me, and their work is checked before it reaches you
- One contract, one invoice, one place to chase