One announcement said a million. The document said a thousand.
Three significant AI announcements landed in the last week of September. I read all three documents rather than the coverage of them, and one had a gap between the headline and the text large enough to be worth an article on its own.
It is a thousandfold gap. And it is the single most useful thing I can teach you about this field.
The case: OpenAI and America's SBDC
On 29 September, OpenAI announced a partnership with America's Small Business Development Center network. Summaries of it described a programme to bring AI to a million small businesses.
Here is what the announcement actually commits to. OpenAI will train around 150 SBDC advisors, and bring hands-on workshops to SBDC networks reaching at least 1,000 small businesses. The workshops are a three-hour format. There are companion guides for marketing, operations, manufacturing, exporting and government contracting.
So where does a million come from? The SBDC network serves roughly a million entrepreneurs each year. That is the size of the network, not the size of the programme. One number describes the room; the other describes how many chairs are being filled.
I am not accusing anyone of lying. This is how a number drifts: a document mentions two figures, a summary keeps the larger one, and by the third retelling the larger one is the headline. It happens faster when the summarising is automated — the news page on this site produced exactly that conflation before I caught it, which is why that page tells you to read the original rather than my summary of it.
The practical consequence is simple. If you had read “a million small businesses are getting AI training” and concluded the market was about to be saturated with AI-literate competitors, you would have made a planning decision on a number that is a thousand times too large.
What OpenAI actually shipped at DevDay 2026
The same day, DevDay produced announcements with far more substance than the SBDC partnership, and considerably less coverage aimed at small business. The ones that matter if you run a company:
GPT-6.1 Sol. OpenAI describes it as offering near-Astra intelligence at a fifth of the standard input and output token prices, available to all users including Plus and Pro. If you have been holding off on AI work because the per-token cost did not justify it, that calculation just changed.
Dots. Always-on agents that work on your behalf between conversations rather than only when prompted. Available on Pro and Business Premium plans in eligible markets. This is the first consumer-facing step toward an assistant that does things while you are not watching.
@ChatGPT in Slack and Microsoft Teams. Mention it in a channel, thread or direct message and it helps with the team's work — without each person needing an individual licence. For a small team this is the lowest-friction AI adoption path yet announced, because nobody has to change where they work.
ChatGPT Space and Sites with Plugins. Shared workspaces where teammates and agents build on the same knowledge, and the ability to add plugins to Sites you build.
Notice which of these is most likely to change your week. It is not the one with the most dramatic name.
What Meta launched
On 28 September Meta announced its Enterprise Platform, described by Mark Zuckerberg as the next major pillar of the business, under CJ Desai as Chief Enterprise Platform Officer. It brings together the Muse agent, Meta Business Agent, Muse API and Muse Code.
This one is worth reading carefully for what it does not yet say. It is a platform announcement and a hiring announcement. What it will cost, what it will do for a ten-person company, and when any of it is generally available are not in the document. That is not a criticism — it is an accurate description of the stage this is at, and the right response is to note it and wait rather than to plan around it.
How to read one of these in four minutes
This is the method, and it works on every announcement in this field.
1. Find the primary document. Not the article about it. The company's own post, the help-centre page, the order, the filing. If coverage does not link to one, be suspicious of the coverage.
2. Find the number, and check what it counts. Almost every inflated claim is a real number attached to the wrong noun. A million entrepreneurs served by a network is not a million businesses trained. Revenue run rate is not revenue. Eligible merchants is not participating merchants.
3. Find who it binds. This decides whether it is your problem. A federal order binding agencies is not a requirement for your business. A platform available to Pro and Business Premium in eligible markets is not available to you if you are on a different plan in a different country.
4. Ask what is missing. No price, no availability date, no eligibility criteria? Then it is a direction of travel, not a thing you can buy. Both are worth knowing. Only one is worth planning around.
Four checks, four minutes. They would have caught the thousandfold gap above in under sixty seconds, because step two is the only one that mattered.
Want to know which of these actually touches your business? The free audit shows how AI assistants currently describe you — and which sources they read to decide.
Run the free auditWhat I would actually do this week
Nothing dramatic, which is usually the right answer.
If your team already lives in Slack or Teams, try @ChatGPT there. It is the one announcement from this week with a near-zero adoption cost, and it will teach you more about where AI genuinely helps your operation than any amount of strategy discussion.
If you previously priced out an AI project and the token cost killed it, price it again against GPT-6.1 Sol. A fifth of the cost changes which projects clear the bar.
If you are near an SBDC and the workshops reach your area, go. A three-hour hands-on session is worth more than a month of reading. Just do not assume a seat is waiting — there are about a thousand of them nationally in this phase.
And if an announcement this week made you feel behind, check the number before you act on the feeling. That is the whole discipline. It is not sophisticated, and almost nobody does it.
Common questions
Is OpenAI training a million small businesses?
No. The announcement of 29 September 2026 commits to training around 150 Small Business Development Center advisors and reaching at least 1,000 small businesses through in-person workshops. The figure of one million refers to the number of entrepreneurs the America's SBDC network serves annually, which is the size of the network rather than the scope of the OpenAI programme.
What was announced at OpenAI DevDay 2026?
The announcements most relevant to businesses were GPT-6.1 Sol, described as offering near-Astra intelligence at a fifth of standard token prices; Dots, always-on agents available on Pro and Business Premium plans in eligible markets; @ChatGPT in Slack and Microsoft Teams, which works without individual licences for each team member; ChatGPT Space for shared team workspaces; and the ability to add plugins to Sites.
What is the Meta Enterprise Platform?
Announced on 28 September 2026, it is a new business division bringing together the Muse agent, Meta Business Agent, Muse API and Muse Code, led by CJ Desai as Chief Enterprise Platform Officer. The announcement does not state pricing, general availability or eligibility, so at this stage it indicates direction rather than something a business can currently buy and deploy.
How do I tell which AI announcements matter for my business?
Read the primary document rather than coverage of it, check what each number actually counts, establish who the announcement binds or who is eligible, and note what is missing. An announcement with no price, no availability date and no eligibility criteria describes a direction of travel rather than something you can act on. Four checks take about four minutes.
Why do AI announcements get exaggerated?
Usually not through dishonesty but through drift. A document contains two real numbers, a summary retains the larger one, and by the third retelling the larger figure has become the headline. Automated summarising accelerates this, because a model optimising for a compelling summary will reach for the bigger number unless something checks it against the source.
Sources
- OpenAI — Helping small businesses put AI to work (SBDC partnership, 29 September 2026)
- OpenAI — DevDay 2026 recap
- Meta — Launching the Meta Enterprise Platform (28 September 2026)
About the author. Siamak Kalhor is an AI marketing and LLMO consultant in Los Angeles with more than forty years of experience in marketing strategy. He advises professionals and e-commerce leaders on becoming the answer AI assistants give. Get in touch, or connect on LinkedIn.