AI tools have become part of how our team works. We use them in content, SEO, automation and design – and we have also seen the problems they create when used carelessly. This article is our honest assessment, based on how AI appears in our projects. We describe it qualitatively; AI is a tool across many projects rather than a separate project type with its own statistics.
Where AI helps
First drafts of website copy
Content is the most common reason website projects stall. WordPress projects with missing or late content had a median delivery time of 74 days in our data, compared with 26 days when content was ready. AI helps us produce a first draft from the client's brief, existing materials and a short interview, which the team then edits.
The rule we never break: every AI-assisted text is reviewed and confirmed by the client. Only the client knows which claims are true, which terms they may use and how they want to sound.
SEO research and on-page work
AI speeds up keyword clustering, meta description drafts, FAQ ideas and content outlines. Research tools still provide the data; AI helps organise and draft. We also add llms.txt files in premium SEO setups, helping AI assistants understand a site's content.
Lead qualification in automations
In our n8n lead generation workflows, an AI step evaluates each lead against the client's ideal customer profile before paid enrichment happens. With clear criteria and structured output, it filters effectively. Borderline cases go to human review.
Call scripts and message drafts
For sales and CRM projects, AI helps draft call scripts, email sequences and follow-up messages, which the client adapts to their tone.
Internal efficiency
Summarising long feedback documents, structuring requirements from workshop notes, drafting test cases – everyday tasks where AI saves time behind the scenes.
Where AI creates problems
Product images
Clients increasingly generate images with AI. In one shop project, AI-generated images showed garments with incorrect brand marks. AI does not reproduce logos and product details reliably. Product and collection images must show the real product; showing something different creates liability. AI imagery can work for mood and storytelling sections.
Claims that are not true
AI writes confidently. If nobody checks, texts can include claims about experience, certifications or services the business does not have. That is why client review is mandatory.
Generic content
AI drafts based on generic briefs sound like everyone else. The more real details a client provides – customers, examples, differences – the better the result.
Legal transparency
EU rules on transparency for AI-generated content are evolving. We have implemented image labelling for AI-generated images on a client site in line with the EU AI Act. Businesses should know where AI content appears on their site.
Automation without oversight
An AI step that decides who gets contacted, without guardrails, can make embarrassing or costly mistakes. We always add thresholds, logging and human review for uncertain cases.
Our rules for AI in client work
| Area | Rule |
|---|---|
| Copy | AI drafts, humans edit, client confirms |
| Product images | Real products only; AI for mood sections |
| Claims and facts | Verified by the client |
| Automation | Clear criteria, thresholds, human review for edge cases |
| Transparency | Label AI-generated content where required |
| Data | No confidential client data in tools without appropriate agreements |
How we decide whether to use AI for a task
A simple test guides us: does AI reduce effort without reducing accuracy, and is there a human check before the result reaches the public or a customer? If both are true – first drafts, research, summaries, internal preparation – AI is a good fit. If the output goes directly to customers without review, or accuracy is critical and hard to verify, we either add a review step or do not use AI for that task.
We also watch costs. AI calls in automations are cheap individually but add up at volume, which is why we place AI steps after cheaper filters in our workflows.
What this means for clients
AI makes some parts of projects faster – especially getting from an empty page to a first draft. It does not remove the need for your input, your review and your decisions. In our data, those human factors – ready content, quick feedback, one decision maker – still determine how fast a project moves.
FAQ
Will my website texts be written by AI?
We may use AI to produce first drafts, which our team edits. You review and confirm every text before launch.
Can AI replace a photo shoot?
For product images, no. For atmospheric images and backgrounds, AI can be an option, with care.
Do I need to label AI content on my website?
It depends on the type of content and current regulations. We advise on labelling where it is relevant.
Why do custom software projects need an out-of-scope list?
Because new ideas appear as soon as people see working software. In our experience, writing down what is not in the first version is what keeps an MVP on schedule.
Why can't you give a fixed timeline before scoping?
Because the timeline depends on rules and integrations that are usually only clear after a scoping session. We prefer a short scoping phase to a guess that later has to be corrected.
Where the numbers come from: our own project records in Odoo – 233 finished projects, 909 tasks and about 4941 logged hours, mostly from 2025–2026. Durations are calendar days from kickoff and we quote medians, so one unusually long project does not distort the picture.
Planning a automation project?
If you want a realistic plan for your own project, book a short call with our team and we will walk you through the phases, what we need from you and where projects like yours usually lose time. You can also browse finished work in our automation portfolio.
Comments