I checked every deployment of AI in the circular economy that I could trace to a primary source: project pages, funder records, conference papers and public procurement notices. The list is shorter and more lopsided than conference talks suggest.
What the register shows
- Public money for circular-economy AI goes to cameras. Of 97 public procurements of AI for environmental work in the UK and Europe since January 2023, 15 touch the circular economy. Eleven of those are sorting robots and contamination checks. None is about reuse or repair.
- Reuse and repair run on pilots and charity shops. Two triage tools for used electrical goods are in testing, one repair assistant has been studied with nine people, and UK charity shops use AI to write online listings from a photo.
- Generative AI does three jobs on this list. It writes listings for donated goods, answers residents’ recycling questions and, in one study, guides a repair. What is bought and running at scale is computer vision.
- Nearly every result is self-reported. The headline numbers come from a single test week, a vendor case study or a projection. I found no independent evaluation of any of them.

Judged by who keeps paying, two uses work today. Recycling operators and councils keep buying cameras and sorting robots, and Biffa and FCC Environment now report AI-measured waste composition to the regulator. Oxfam GB moved its listing tool from pilot to an enterprise contract. Everything else on this page is still being tested.
The register covers the field end of the circular economy (collection, recycling, reuse and repair), where community organisations and small consultancies work. AI in product and material design happens inside manufacturers, so it is left out.
I have spent ten years on circular-economy projects in the field, from a coastal city waste system in Indonesia to a fishing-net recovery loop in the Seychelles that became my field methodology. I use AI every day for the desk side of that work. This page is the register I keep of where AI is doing the field side, with what each system does, its status and a source you can check.
In this hub
- Reuse and repair, eight deployments from intake to resale
- Recycling and collection, what public bodies buy
- Residents’ questions, chatbots that say which bin to use
- What AI does not do, four limits in the same sources
- Where AI helps a small team, and where it does not yet
- Azvai’s own evidence, field tests and open data
Reuse and repair run on pilots, a study and charity shops
This is the end of the circular economy that keeps products in use, and the end where community organisations work. Everything I could verify is in the table.
| Deployment | What the AI does | Status | Source |
|---|---|---|---|
| Circulus triage tool, Apeldoorn (Netherlands), Interreg project E6 | Identifies a used electrical device from one photo at intake, so staff can decide faster whether it could be reused | Pilot. One test week reported in August 2026 | E6 project news |
| RecyclingBörse, Herford (Germany) | App to recognise small electrical appliances and show repairability and likely resale value | In development since early 2024, funded by the German Federal Ministry of Economics | RREUSE member story |
| Oxfam GB with Thriftify (UK) | Turns a photo of a donated item into a marketplace listing and posts it to several sales channels | Live. Enterprise agreement in August 2026, after a pilot | UK Fundraising |
| British Heart Foundation (UK) | Suggests a price and a sales channel (shop, eBay or Depop) for donated items | Live in some shops since January 2025 | Microsoft case study |
| Midlands Air Ambulance Charity with Minimist (UK) | Writes eBay and Vinted listings with a suggested price from one photo | Rolling out to 11 shops in 2026. No results published | Charity announcement |
| Wargön Innovation and RISE (Sweden) | Open image database to describe, value and sort second-hand clothing | Pilot project due to end in April 2024. I found no later report of use | Project press release |
| AI-Fixer, Royal College of Art (UK) | Chatbot that guides a person step by step through a phone screen or headphone battery replacement | Research study presented at PLATE 2025 | Conference paper |
| REEP, University of Galway (Ireland) | Proposed an AI marketplace to match reuse facilities, repair groups and citizens | Concept. Science Foundation Ireland lists the challenge entry as complete, and I found no public product | SFI challenge page |
Triage of used electricals
The Circulus test is the clearest description of AI at a reuse intake desk. Circulus, the waste company for Apeldoorn, ran a test week with Foenix Circulair Ambachtscentrum and the Belgian research institute VITO.
257
devices registered with the tool in the test week
40%
flagged as potentially reusable
92%
identified from a single photo, as the project reports
Those are first results from one week, and the post does not say how accuracy was measured. The pilot sits inside E6, a project on extending the life of electrical equipment under Interreg North-West Europe, an EU programme that funds cross-border projects. E6 runs from January 2024 to June 2027 with about €4.9m of EU funding (keep.eu).
RecyclingBörse, a social enterprise in Herford and a member of the European reuse network RREUSE, handles small electrical appliances for about 250,000 residents. It is building a similar app with a university, a software company and repair and recycling partners. Until it is finished, staff keep assessing items from photos.
Both projects aim AI at the same moment, the few seconds when someone decides whether an item goes to the shelf, the bench or the skip.
Listings for charity shops

The most mature reuse deployments are in UK charity retail, and the listing tools among them are generative AI.
292%
more listings per lister per day in Oxfam’s pilot
24%
higher average sale price in the same pilot
Nearly 800,000
items a week across British Heart Foundation shops, by the charity’s count
11
Midlands Air Ambulance shops getting Minimist in 2026
- Oxfam GB signed an enterprise agreement with Thriftify in August 2026 after a pilot. Its head of e-commerce says the platform lets Oxfam list more items and sell them at a higher average price. The two pilot figures are the pilot’s own numbers, and the coverage credits the second to Thriftify’s own case study.
- The British Heart Foundation built its own tool on Microsoft’s platform and has used it in some shops since January 2025. The £1m in extra sales in the case study headline is a projection in the body text, and the software vendor published the case study.
- Midlands Air Ambulance Charity is rolling out Minimist and has not published results yet.
Listing suits AI for a plain reason. It is repetitive, it happens at volume and a person checks every result before it goes live, so a wrong draft costs a minute and a good one saves ten.
Repair guidance
AI-Fixer is the one verified study of generative AI helping people repair, presented at PLATE, the main academic conference on product lifetimes. Nazli Terzioglu at the Royal College of Art ran 16 co-design workshops with nine participants, five with no repair experience and four experts. They replaced a Fairphone 5 screen and a FairBuds XL battery.

- What helped. Non-experts found the chatbot a helpful entry point, and one participant moved on to a harder phone repair after finishing the headphones.
- Where it fell short. Text alone was not enough to find the right cable connector or screw, so novices asked for images and voice. Experts found the answers long and repetitive.
It is a feasibility study with a handful of people, and it is honest about that.
What public bodies buy for recycling and collection
For my analysis of who buys AI for environmental work, I started from every public procurement notice that mentioned AI in the UK and Europe from January 2023 to October 2026, on the UK portals and on TED (the EU’s official tender journal).
None of the 97 is about reuse or repair, and neither is any of the 38 borderline cases I also kept. You can filter every procurement yourself in the tender explorer.
Below is a selection, with one long-running UK industry deployment added. Notices record what was bought, so none of them tells you whether purity actually rose. Each row cites its official notice number, and the open CSV in the evidence section links every notice.
| Buyer or operator | What the AI does | Status | Source |
|---|---|---|---|
| KOCEROD waste company (Slovenia) | AI and machine-vision robot on a packaging sorting line | Bought in 2023, €348,968 | TED notice 283202-2023 |
| SIA ZAAO (Latvia) | Sorting plant with an AI-guided robotic grab | Bought in 2026 | TED notice 617104-2026 |
| Métropole Toulon-Provence-Méditerranée (France) | Cameras on collection trucks that check sorting quality | Bought in 2026, €103,600 | TED notice 91843-2026 |
| Grand Besançon Métropole (France) | Truck sensors, including AI-assisted detection of sorting errors | Tender open on 20 October 2026, €2.1m | TED notice 470981-2026 |
| West London Waste Authority (UK) | Cameras on collection vehicles to spot recycling contamination | Market engagement in 2026 | Find a Tender notice 079773-2026 |
| Greater Manchester Combined Authority (UK) | New dry-recycling sorting plant in Bolton, with bidders asked to use AI to raise capture and purity | Tendered in 2025, £20m, outcome not published | Find a Tender notice 022087-2025 |
| Biffa and FCC Environment (UK) | Greyparrot cameras measure waste composition on conveyor belts | Live for several years. AI-derived composition data submitted to the Environment Agency, reported in May 2026 | letsrecycle.com |
Germany adds a funding programme. The federal environment ministry, through its agency ZUG, runs the Green-AI Hub Mittelstand, which develops AI for material savings in small and mid-sized manufacturers. Its follow-on framework went to the German Research Center for Artificial Intelligence in 2026 for €19.6m (TED notice 102964-2026).

Everything in this table checks what people have already thrown away. It raises the purity of recycled material and catches the wrong item in the wrong bin.
None of it keeps a product in use for longer. None of the 97 procurements or 38 borderline cases buys AI for reuse, repair, refurbishment or second-hand goods, in any language.
Councils do pay for reuse and repair, with no AI involved. In Scotland, 40 public procurements since 2023 bought reuse or repair, and 83 more bought the handling of a waste stream where reuse was possible without asking for it.
Answering residents’ questions
The third cluster is generative AI facing the public, mostly about which bin to use.
| Deployment | What the AI does | Status | Source |
|---|---|---|---|
| WRAP, for a Hampshire council pilot (UK) | WhatsApp chatbot on food waste and recycling | Bought in 2025 | Find a Tender notice 085430-2025 |
| Balearic Islands government (Spain) | AI assistant on the web portal of the circular economy and climate directorate | Bought in 2025, €27,165 | TED notice 499501-2025 |
| Garbiñe AI (Basque Country, Spain) | Says which bin an item goes in, by text, voice or photo, and finds the nearest container from Basque open data | Built by Beatriz Arenal Redondo, won the general category of the fifth Open Data Euskadi awards in 2024. The app did not load when I checked on 8 November 2026, so it may be offline | datos.gob.es listing |
Garbiñe came from one person working with open data. The two public purchases are recent, and neither notice reports how residents use the service.
What AI does not do in this work
The same sources show four limits, for generative AI and the older systems alike.
1. It does not make the reuse decision.
Circulus flags items as potentially reusable, and Herford’s staff still assess from photos. Testing, safety checks and the final call stay with people at the bench.
2. It does not replace hands and eyes in a repair.
AI-Fixer’s novices could not find the right connector from text, and the experts wanted shorter, adaptive answers. A chatbot gets someone started. It does not see the device.
3. It does not create a market.
Sysav built Sweden’s automated textile sorting plant, Siptex, and announced in December 2025 that it would sell it because the market for sorted textiles had not grown as expected (EUWID). That plant used no generative AI, and the lesson carries over. Better sorting does not create buyers for what comes out.
4. It does not produce its own evidence.
Every number in the reuse table comes from the project, the vendor or a projection. If you are asked to fund or adopt one of these tools, ask how the accuracy or sales uplift was measured, and on what sample.
Where AI helps a small circular-economy team today
For a small consultancy or a community reuse organisation, the register points to the same place my own practice does.
The field work stays human
- walking the site
- running the audit
- training the team
- sitting with the council
AI compresses the screen work around it
- market scans
- policy comparisons
- funder reports
- data analysis
I set out that division in what an AI-augmented consultant actually does, and the workflows I use are in Claude Code for circular-economy research. Anthropic’s own usage data shows sustainability tasks using three to nine times less AI than equivalent work.
If the footprint of the tools themselves is your worry, is generative AI sustainable? covers what the labs disclose and what remains unmeasured.
The one field task where generative AI is already earning money for reuse organisations is the listing. If you run a reuse shop with an online channel, that is the pilot worth watching, ideally with your own before-and-after numbers.
Azvai’s own evidence on AI and the circular economy
This register sits at the centre of a set of pieces built on real or public data. The field tests check the claims in the AI-augmented consultant article one at a time, reporting what came out and what broke. New field tests and data pieces are added here when they are published.
Field test 1
Can AI do recycling market research? I re-ran a 2023 European fishing-net study. 32.5 hours of desk research became about 15 minutes, and the checks caught the AI’s errors.
Open data
All 97 AI procurements as a CSV, with a link to every official notice behind the recycling table.
Survey data
Why Europeans repair uses answers from 27,498 people in 28 countries. Among people otherwise alike, difficulty paying bills makes no detectable difference to who repairs.
Official statistics
Scotland’s household recycling, council by council. The rate peaked in 2017, and the legal 60% target for 2020 was missed by a wide margin.
System map
Scotland’s circular economy system map shows who funds, regulates and delivers circular-economy work in Scotland.
Field case
Scaling a take-back programme across European markets, a comparative case from Interface ReEntry in the UK, Ireland, Spain and Benelux.
How this register is built
A deployment is listed when a primary source shows that it exists: a project or funder page, a conference paper, an official procurement notice or the organisation’s own announcement. Vendor case studies are marked as such. Each status reflects what the source said when I checked it on 8 November 2026.
“Every deployment” means every one I could find in project databases, funder records, conference proceedings, procurement portals and sector news, so a missing project may simply be one I have not found. Candidates I could not source are left out. A Greek “AI-powered” reuse-exchange ecosystem, for example, appears only as a project idea on a matchmaking platform.
The procurement rows come from every AI notice on Find a Tender, Contracts Finder, Public Contracts Scotland and TED. A smaller AI model coded the notices on environmental topics against a written codebook, a larger model reviewed every unclear case and audited a random sample, and the method is set out in the tender article. The circular-economy label was a yes or no call on short descriptions, so treat the 15 as directional. The absence of reuse and repair is firm, because every notice in the topic pool was coded.
If you know of a deployment with a primary source that belongs here, write to borja@azvai.com and I will check it.
If you run a reuse, repair or circular-economy project and want to work out where AI could take the screen work off your team, that is what my AI for sustainability research service is for.
FAQ
How is AI used in the circular economy?
Mostly at the recycling end. Sorting robots and cameras on conveyor belts and collection trucks check material purity and contamination, and public bodies in the UK and Europe buy them. At the reuse and repair end, the verified uses are triage pilots for used electrical goods, charity shops that generate online listings from photos, and a repair-guidance chatbot tested in a small study.
Is generative AI being used in reuse and repair?
Yes, in a few places. Oxfam GB, the British Heart Foundation and Midlands Air Ambulance Charity use AI tools to list or price donated goods. The Royal College of Art tested a repair chatbot with nine people. No public body in our dataset has bought generative AI for reuse or repair, and the published results are self-reported.
Which AI tools exist for repair cafés?
I found no AI tool in use by repair cafés with a primary source behind it. The closest evidence is the AI-Fixer study, where people with no repair experience found step-by-step chatbot guidance a helpful start, but needed images or voice to identify parts.
Where does AI help a small circular-economy consultancy?
In the desk work around field projects, such as research synthesis, market and policy scans, data analysis and funder reports. The field work itself, including audits, stakeholder meetings and training, gets little help from current tools.
