The verdict in three sentences
For a technical supplies distributor in Dublin that needs to enrich 12,000 product pages, human copywriting costs EUR 8 to 15 per item, or EUR 96,000 to 180,000. A custom generative AI pipeline connected to the PIM brings the marginal cost down to EUR 0.10 to 0.50 per item, plus a setup of EUR 8,000 to 20,000. Success depends on one word: control, with technical data taken from the PIM only and human review on a 10% sample.
What a product page costs in 2026: human, hybrid or AI
Thin pages (SKU, truncated name, one photo) hurt both organic search and the conversion rate of trade buyers, who compare specifications before ordering. Here is the 2026 order of magnitude for 12,000 items.
| Approach | Cost per item | Setup | Total for 12,000 items | Timeline |
|---|---|---|---|---|
| Freelance copywriters | EUR 8 to 15 | EUR 0 | EUR 96,000 to 180,000 | 6 to 9 months |
| Content agency | EUR 12 to 20 | EUR 2,000 | EUR 146,000 to 242,000 | 5 to 8 months |
| Consumer AI tool, copy-paste | EUR 1 to 3 (internal time) | EUR 0 | EUR 12,000 to 36,000 | 4 to 6 months |
| Custom AI pipeline + 10% review | EUR 0.10 to 0.50 | EUR 8,000 to 20,000 | EUR 9,200 to 26,000 | 6 weeks |
| Hybrid: AI + human copy for top 500 | EUR 0.50 + EUR 12 on 500 | EUR 10,000 | EUR 21,000 to 26,000 | 8 weeks |
The AI pipeline cost per item includes model calls, storage and sample review. The consumer tool looks cheap, but internal copy-paste and correction time quickly exceeds 5 minutes per item, with no source traceability.
Pipeline architecture: the PIM as the single source of truth
The main risk of generative AI on a technical catalogue is fabrication: a voltage, a diameter or a standard that does not exist. The safeguard is to forbid the model from using any data that does not come from the PIM or the manufacturer sheet.
| Step | Role | Typical tool | Share of budget |
|---|---|---|---|
| PIM extraction | Attributes, units, standards, photos | Akeneo, Pimcore or ERP export | 15% |
| Normalisation | Units, synonyms, templates per family | Custom script | 15% |
| Generation | Title, description, benefits, product FAQ | Language model via API | 25% |
| Automated checks | Numbers present in PIM, length, banned words | Rules + second verifier model | 20% |
| Human review | 10% sample per family | Validation interface | 15% |
| Publishing | Back to PIM, then to the site | PIM to e-commerce connector | 10% |
The automated check compares every number in the generated page with the source attributes. Any value missing from the PIM blocks the item and sends it to review. In our projects, this filter catches 2 to 5% of items on the first pass.
Quality control: thresholds and KPIs
| Indicator | Acceptance threshold | Action if exceeded |
|---|---|---|
| Technical error on sample | under 1% | Fix the family template, regenerate |
| Items blocked by automated check | under 5% | Enrich the PIM |
| Duplicate text across items | under 3% | Vary the templates |
| Description length | 120 to 300 words | Adjust the prompt |
| Review time per item | 1 to 2 minutes | More guided interface |
| Conversion rate of enriched pages | +10 to 25% at 3 months | A/B test on one family |
We always start with a pilot family of 300 to 500 SKUs, measured before rolling out.
Mini case study
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
Claire, e-commerce director at an electrical supplies distributor in Dublin, manages 12,000 SKUs, 9,000 of them thin. Freelance quote: 9,000 × EUR 10 = EUR 90,000 over 7 months. AI pipeline: setup EUR 14,000, generation and review 9,000 × EUR 0.35 = EUR 3,150, so EUR 17,150 in 6 weeks. Savings: EUR 72,850. If enriched pages generate EUR 4,000 of extra margin per month, the project pays back in just over 4 months, and the pipeline then handles every new SKU at EUR 0.35.
FAQ
Does Google penalise AI-generated product pages?
No, Google assesses usefulness, not how content is produced. A 150 to 300 word page built on accurate attributes and 97% unique outperforms a 10-word product name.
Do we need a PIM to start?
A full PIM is not mandatory, a clean ERP export with attributes per family is enough to begin. Allow 2 to 3 weeks of normalisation if fewer than 60% of attributes are filled in.
Who is liable for a published technical error?
The distributor remains responsible for displayed information. That is why the pipeline blocks any value absent from the PIM and enforces a 10% review per family, targeting fewer than 1% errors.
What does maintenance cost after launch?
Budget EUR 300 to 800 per month for hosting, model calls and template tuning. Each new item then costs EUR 0.10 to 0.50.
Can we also generate French and German versions?
Yes, translation adds about EUR 0.05 to 0.15 per item and per language. A native review on 5% of the sample is still recommended for export markets.
Let's scope your project. Send us your SKU count, a PIM or ERP export and your priority families: we will price the AI pipeline, a 500-item pilot and a firm timeline. Detailed quote within 48 h. WhatsApp +221 77 596 93 33.
Mohamed Bah
Fondateur, Kolonell
Passionate about digital and entrepreneurship in Africa, Mohamed has been helping Sénégalese businesses with their digital transformation since 2020. Founder of Kolonell, he believes every SME deserves a professional and accessible online présence.
