What "OpenAI Select Partner" actually means, and how to vet any AI partner's credentials

An agency has pitched you an AI project, and somewhere around slide three there was a badge. OpenAI Select Partner, or something close to it. It did what badges are supposed to do, which is to make the rest of the deck feel safer.

SpotDev holds that badge. We were confirmed as an OpenAI Select Partner on 07 Aug 2026, and we display it on our own site. So take the following as coming from someone with an obvious interest in the credential looking impressive, telling you that it is weaker evidence than it looks.

The badge proves enrolment at a named tier, and that is all it proves

The OpenAI Partner Network launched on 14 Jun 2026. OpenAI describes it in one sentence: "a new program for partners around the world to build, sell, and deliver AI solutions with OpenAI." Partners progress through three tiers, Select, Advanced and Elite, which OpenAI says are gated on sales performance, technical capability, co-sell engagement and deployment experience.

That is the entire public description. A firm displaying the badge has enrolled in a real programme run by OpenAI, at a named tier, and has met whatever internal bar OpenAI applies. That is a genuine fact and it is not nothing. It is also considerably less than most people reading it assume.

OpenAI has not published the thresholds, so nobody can tell you what bar was cleared

We looked for a tier comparison, a benefits table, minimum revenue, deployment counts or certification counts. None of it is published. OpenAI's newsroom post and its partner page name the four gating dimensions and stop there. Trade press covering the launch reached the same conclusion.

That has an awkward consequence for anyone selling on the strength of the badge. We cannot tell you what our Select status required, because OpenAI has not said, and any partner who gives you a confident account of the thresholds is describing something they have not seen either.

What the badge does not mean

Worth being blunt about, because these are the inferences buyers make without noticing:

• OpenAI has not reviewed the partner's work. No inspection of code, architecture or delivery. OpenAI has not looked at a system we built, and it has not looked at anyone else's either.
• It is not an endorsement or a guarantee. OpenAI's own brand rules tell partners not to imply endorsement or misrepresent the relationship. If a supplier's deck reads as though OpenAI vouches for their delivery, that is the supplier's framing, not OpenAI's position.
• It confers no influence over roadmap, support or pricing. When OpenAI deprecates something, it lands on the partner the same day it lands on you. Nobody at Select tier gets a queue-jump on support or a discount to pass on.
• It is not a technical certification of a person or a build. It attaches to the firm, not to the engineer assigned to your project and not to the system they will hand you.
• It does not put us in the Partner Locator. We are not listed in OpenAI's partner directory. If a supplier points you at that directory as their proof, note what it actually evidences: a position in a vendor's programme, not whether they can build the thing you need.

Where Select sits in the structure

Tier is not a proxy for company size. OpenAI's gating factors are all about activity with OpenAI specifically, which is why the Select tier contains a mix of boutique AI specialists and some very large firms newer to the programme.

We are at the entry tier. Anyone who reads "OpenAI Select Partner" as the top of a ladder has read it wrong, and any supplier who lets that impression stand is trading on the ambiguity.

The credential is heading for table stakes, quickly

OpenAI has stated a target of training and enabling 300,000 certified consultants by the end of 2026. The Partner Locator listed 60 partner firms when we checked it on 08 Aug 2026.

Hold those two numbers next to each other. A credential moving towards six figures of certified people inside eighteen months cannot stay a differentiator for long. Within a year, "we are an OpenAI partner" is likely to be closer to "we have a company registration number" than to a mark of distinction.

So the badge is useful for exactly one thing: sorting the pile. It tells you a supplier enrolled in the vendor's programme. It does not tell you they can build what you need.

Nine questions to ask any AI supplier, including us

None of these require technical knowledge to ask, and all of them are hard to bluff.

1. What have you actually shipped, to whom, and who operates it now? You are looking for a named system in production, with a real operator. "We ran a pilot" and "we delivered a strategy" are different answers to a different question. Ask who is running it today and whether you can speak to them.

2. Who owns the code, and whose name is on the model contract? The failure mode is a supplier who builds on their platform, against their OpenAI account, and hands you a login. It works fine until you want to change supplier, at which point you discover you own nothing. The answer you want is that the code sits in your repository and the vendor contract is in your company's name.

3. Do you resell licences, or take a margin on usage? If your supplier earns more when you consume more tokens or move to a higher tier, their advice on architecture and tier selection is compromised. It does not make them dishonest, it makes them conflicted, and you should price that in. For the record, we resell nothing and take no margin on usage.

4. What have you had to migrate since 01 Jan 2026? This is the question that separates firms who have shipped from firms who have presented. The 2026 deprecation schedule gave everyone plenty to answer with. The Assistants API retires on 26 Aug 2026. DALL-E was shut down on 12 May 2026 and replaced by GPT Image 2. GPT-5.3 through 5.5 have moved to Legacy. AgentKit's Agent Builder and Evals are winding down, with Evals going read-only on 31 Oct 2026.

A supplier with production systems will have a specific, slightly weary answer about what they moved and when. A supplier who cannot name a single migration either has nothing in production or was not paying attention. Both are disqualifying.

5. Which controls does the tier you are recommending not include? Ask this before you sign, not during your first security review. The gap between ChatGPT Business and ChatGPT Enterprise is where most unpleasant surprises live. From OpenAI's own comparison table, checked 08 Aug 2026:

Two details that catch people out. Data residency applies only to new Enterprise or Edu workspaces and to new API projects, and cannot be added to an existing one. Zero Data Retention is approved through OpenAI's sales process rather than switched on in a settings panel, and it carries carve-outs for safety review. A supplier who describes either as a checkbox has not configured one.

6. Who is my data processing agreement with, and on what transfer mechanism? For a UK customer the counterparty under OpenAI's DPA is OpenAI OpCo, LLC in the United States, and UK data transfers run on the EU standard contractual clauses as amended by the ICO's UK Addendum. You do not need your supplier to give you a legal opinion, and you should be wary of one who offers it. You do need them to know who you are contracting with, because it goes in your DPIA and a supplier who has never opened the document will not know.

7. What are the training defaults on what you are proposing? Business and API data is excluded from model training by default. Consumer ChatGPT is included by default and has to be actively switched off. Suppliers invert this constantly, in both directions, and getting it wrong in either direction tells you they are working from memory rather than from the documentation.

8. Can I meet the engineer who will build it? Not the account director, and not the solutions architect who presents at the pitch and is never seen again. Ask who writes the code, ask to speak to them before you sign, and ask what proportion of their time your project gets.

9. What would make you tell us not to build this? A supplier with no answer has never turned work down. Sometimes the correct recommendation is to use the AI features already included in software you pay for, or to fix the underlying data first, or to do nothing this quarter. If your supplier's answer to every problem is a build, you are buying their business model rather than their judgement.

Apply this to us

We hold the OpenAI Select Partner badge at the entry tier, and we have just spent a page explaining why that should not move your decision much. We are also a HubSpot Diamond partner, Cyber Essentials Plus certified, and rated 5.0 from 38 reviews, and none of those should carry the decision on their own either.

So put us through the checklist. Ask what we have shipped and to whom. Ask who owns the code. Ask what we have migrated since 01 Jan 2026. Ask us which model we would recommend and why, on OpenAI work, on Claude, or on whichever platform actually fits, since choosing between them is the point of an independent AI implementation partner.

If you are earlier than this and still weighing the options, start with what ChatGPT and the OpenAI API actually do for a UK business.

The decision in front of you is not which supplier has the most badges. It is which supplier can show you a system running in production, name what broke and what they did about it, and tell you when the answer is not to buy anything. Ask the nine questions of every firm on your shortlist. If you want our answers, request a quote and we will give them to you in writing.