Content at scale is a data problem before it is a writing problem. The reason product descriptions are inconsistent is usually that the underlying records are inconsistent, so the first job is structure: what facts exist, which are reliable, and which are missing.
From there we build a generation pipeline that writes from those facts, in a defined voice, with the specifics that matter for search. Measurements, materials, scale, compatibility, service area. The details buyers search for and generic copy always omits.
Nothing goes live unreviewed. Output lands in a queue with the source facts alongside it, so a person can approve a batch in minutes instead of writing for weeks.
What is included
Content model
A defined structure for what each piece of content must contain, so every item is comparable and complete.
Voice specification
A written brand voice with worked examples and explicit prohibitions, used as the instruction set for generation.
Generation pipeline
Batch generation from your catalogue or data source, with retries, duplicate detection and cost controls.
Review workflow
An approval queue showing generated copy beside its source facts, with edit, approve and reject in one pass.
Search optimisation
Titles, meta descriptions and headings generated to a template that respects length limits and keyword intent.
Publishing integration
Approved content pushed into your CMS or store rather than exported to a spreadsheet someone has to paste from.
How we work
- 01
Audit the catalogue
We measure what is actually missing. On one recent catalogue of more than three thousand products, only fourteen mentioned scale, which is the single most searched attribute in that market.
- 02
Define voice and template
We write the specification and generate a sample batch for you to mark up, before anything runs at volume.
- 03
Generate and review
Batches run, land in review, and get approved. Rejections feed back into the prompt so quality climbs across the run.
- 04
Publish and monitor
Approved copy publishes, and we track indexing and search performance so the work can be judged on results.
Best suited to
- E-commerce catalogues where most products have thin or duplicated descriptions
- Multi-location or multi-service businesses needing genuinely distinct pages, not spun templates
- Ongoing publishing programmes that keep stalling because writing is the bottleneck
- Marketplace and listing businesses where every item needs copy the day it arrives
The line we will not cross
We do not publish unreviewed content, and we do not generate near-identical pages to game a search engine. Both damage the site that hosts them.
The value here is producing accurate, specific, genuinely useful content at a volume that manual writing cannot reach. Anything else is a short-term trick with a long-term cost.
Frequently asked questions
Will Google penalise AI-written content?
Google’s guidance targets low-value, unhelpful content regardless of how it was produced. Accurate, specific, reviewed content built from real product data performs well. Mass-produced filler does not, whoever writes it.
Does it sound like our business?
It sounds like the specification we write together. That specification is built from your existing best-performing copy and refined against a sample batch before any volume run.
Can it handle thousands of items?
Yes. Batch processing, cost controls and duplicate detection are part of the build precisely because volume is the point.
Who reviews the output?
Usually your team, in the queue we build. If you would rather we handled review, we can quote that as a separate line.
What if our product data is incomplete?
Then the audit will show it, and we fix the data first. Generating confident copy from missing facts is how you end up with plausible, wrong descriptions.
Related work
Auction House
A bespoke AI-assisted description generator for a busy auction house, staff upload item photos, a vision-language model drafts a professional listing in seconds, and experienced staff review before publication, combining AI speed with human quality control.
Read the case study →Next projectYours
We are taking on new AI Content Systems work now. Tell us what you are trying to fix and we will tell you honestly whether we are the right people for it.
Start the conversation →Often paired with
Or see every service we offer.