OpenAI Dots vs Grok Bot: an honest comparison for businesses
Two agents, one choice. I use Grok Bot, through Cursor Pro+, from the Netherlands. OpenAI's Dots I cannot try: the personal Pro licence is closed in the EEA and only Business Premium works. So for most Dutch business owners Grok Bot is the agent you can start with today. Dots fits a company that already has a ChatGPT Business workspace.
Say you run a business (an example, not a real person). You have two tabs open. One says Dots, the other says Grok Bot. Both promise an agent that keeps working while you sleep and comes back when it needs your approval. Which one fits you? I am in a strange position: I use one and cannot even try the other. I keep saying so, because it decides how sure I can be.
Updated on 15 October 2026. Dots is two weeks old and Grok Bot nine. The reviews I read are one to two days old, and no independent test of agent tasks exists. Where the evidence is thin, I say so. This is part three of a series. Part one covers OpenAI Dots, part two covers Grok Bot.
Dots and Grok Bot side by side
First the facts in one place, as of 15 October 2026. If a cell says nothing, the vendor publishes nothing.
| Criterion | OpenAI Dots | Grok Bot (SpaceXAI) |
|---|---|---|
| Provider | OpenAI | SpaceXAI (formerly xAI), running on Cursor infrastructure |
| Announced | 29 September 2026 at OpenAI DevDay | 11 August 2026, as a paid beta |
| Model | GPT-6 Astra, according to OpenAI | No model picker. Cursor chooses the models. On 6 or 7 October Musk wrote on X that the best model will be used per task |
| How the computer works | Every dot has its own cloud computer and browser | One persistent Linux computer per user. According to SpaceXAI docs, all Bots of one user share it, including files, browser sessions and logins |
| Price and route to access | First dot at no extra cost with ChatGPT Pro and Business Premium. Not on Free, Go or Plus. Current prices: chatgpt.com/pricing | Included in SuperGrok, Cursor Pro and Teams plans with a separate usage pool. Cursor lists it from Premium upward. Also in The X Subscription since 12 October, tier unspecified. Current prices: cursor.com/pricing and x.ai/news/x-subscription |
| Free trial | None found | One-off usage credit for accounts on a free Cursor plan, with a 7-day window. Size of the credit unpublished |
| Availability in the Netherlands and EU | Pro: 18 and older, except the EEA, Switzerland and the UK. Business Premium: all supported ChatGPT regions. Enterprise, Edu and Healthcare: beta, only if the admin turns it on | No official region list. One secondary source (ai.nl, 14 September) lists the Netherlands as available with a paid plan. I use it from the Netherlands via Cursor Pro+ |
| Apps and connections | More than 4,000 apps through plugins, including Slack, Teams, Google Workspace, GitHub, Figma, Outlook | Connectors for Gmail, Google Calendar, Drive, Outlook, Teams, SharePoint and Salesforce. In apps without an API the Bot logs in and operates the interface |
| Autonomy rules and approvals | Custom Rules with four behaviours (act without asking, act if pre-approved, ask first, hand over). Auto-review of risky actions. Activity View and a pause button | Allow once, Always allow, Deny. Optional Auto Review. Background approvals expire after about 10 minutes. Passwords, 2FA, CAPTCHA and payments you take over yourself |
| Where the computers run and data residency | No data or inference residency in the enterprise beta, according to a secondary source | Computers run in the United States. Contractual residency through the account team |
| Channels | ChatGPT on desktop, web and mobile, plus Slack and Teams. SMS contradictory | Apps for macOS, Windows and iOS 18 or later. No email, Slack or WhatsApp channel |
| Strengths according to testers | Content plans (9.8 out of 10), a branded site (9.6), calendar management | A weekly tracking routine that ran every Friday, tools without an API (8 out of 10), draft-first knowledge work (7) |
| Weaknesses according to testers | An email sent unnoticed, unrequested proactive actions, slow, failed phone call, broken spreadsheet formulas, inbox triage (7.5) | Fast usage, fragile browser work (CAPTCHA, changing pages), support queue (4 out of 10), a lead scout that delivered nothing in week one |
| Sources | OpenAI announcement and help centre, TechCrunch, NBC | x.ai announcement and docs, Cursor help pages, The Next Web, Gizmodo |
Read the scores as stories from single reviewers.
How do you get in, and where are the prices?
I name no amounts in this article. The prices of both products changed within weeks, and older articles are already wrong. The price the press quoted at the Grok Bot launch in August is outdated. Look at the vendors' own pages: chatgpt.com/pricing for Dots, and cursor.com/pricing and x.ai/news/x-subscription for Grok Bot.
I can explain the routes. Grok Bot has had no separate price since 26 August 2026. It sits in all SuperGrok, Cursor Pro and Teams plans, with a usage pool of its own. Cursor lists it from Premium upward, which I read on 15 October. The X Subscription, announced on 12 October, bundles Grok, Grok Bot, Cursor and X Premium, and the announcement does not say which tier contains Grok Bot. Accounts on a free Cursor plan also get a one-off trial credit with a 7-day window, according to Cursor's help page. That lets you test without paying.
With Dots, the first dot comes at no extra cost with ChatGPT Pro and Business Premium. Pro is closed to the EEA, so Business Premium is what remains. OpenAI names no amount for it on its price page. Secondary sources disagree, and one reports a minimum of two seats.
Neither vendor publishes limits per plan. OpenAI says conversations with a dot do not count against ChatGPT limits, while Codex and Work tasks do. Nobody knows what a task costs. Without amounts and limits I will not claim which is cheaper. The difference in route is clear: Grok Bot has a trial and several entry points, and Dots needs a ChatGPT Business workspace in the EEA.
Can I use them in the Netherlands?
With Dots it depends on your subscription. Business Premium works in all supported ChatGPT regions, and Enterprise, Edu and Healthcare have a beta that an administrator has to switch on. The personal Pro licence does not exist in the EEA, Switzerland and the UK. OpenAI gives no reason and no date. It only says more users are coming soon.
Grok Bot has no official region list. I found no primary evidence that the Netherlands is blocked or explicitly supported. ai.nl lists the Netherlands as available with a paid plan, and I use it myself from the Netherlands through Cursor Pro+. That is one data point about access. It says nothing about quality. So a Dutch freelancer can start with Grok Bot today and cannot start with Dots Pro. A company with a ChatGPT Business workspace can start with Dots.
I have not used Dots, because the Pro licence is closed to me. What I write about Dots comes from OpenAI documents and reviews by others.
How safe are they?
Both vendors say plainly that prompt injection is not solved. In that attack, a web page or email contains hidden instructions for the agent. OpenAI describes defences and monitoring that can pause or stop a dot. SpaceXAI's docs say its defences do not remove the risk and that heavy work sits behind approval.
One difference matters for a small business. With Grok Bot, all Bots of one user share a single computer, with the same files, browser sessions and logins. The docs themselves say separate Bots are no security boundary. Run one Bot for marketing and one for invoices, and both reach the same logins. With Dots every dot has its own cloud computer. Linking your own computer is optional and off by default. A dot has no email address of its own at launch.
Then the data. Grok Bot computers run in the United States. You transfer data to a country outside the EU, which means a data processing agreement and probably a data protection impact assessment. Grok Bot requires data storage and does not work with Privacy Mode (Legacy). Files are deleted or returned within 30 days after the end of service. For Dots, the enterprise beta has no data or inference residency, according to a secondary source, so nobody guarantees your data stays in the EU. OpenAI does not use data of Business, Enterprise and Edu for model training by default.
The track record counts too. Reviewers of Dots report an email that went out unnoticed and actions nobody asked for, and the live demo failed during the keynote, according to NBC. For Grok, the European Commission opened a Digital Services Act investigation into Grok on X on 26 January 2026. I found no final fine. Liability under the Grok Bot terms is capped at the larger of the fees paid or a fixed minimum. Musk said on X that SpaceXAI would make you whole if the Bot messes up, but that sentence is absent from the terms. On paper Dots has the safer architecture, with a computer per dot. Neither gives a legal statement about the GDPR. Have a lawyer look before real customer data goes in.
Does it work with my tools?
Dots has the longer list: more than 4,000 apps through plugins. OpenAI names Slack, Teams, Google Drive, Gmail, Calendar, Docs and Sheets, GitHub, Figma, Adobe Creative Cloud, Outlook and Codex. It publishes no complete list. Grok Bot names fewer connectors: Gmail, Google Calendar and Drive, OneDrive, Outlook, Teams, SharePoint and Salesforce, with Google Docs, Sheets and Slides added on 22 September. If an app has no API, the Bot just logs in and operates the interface, the way you would. Testers call that fragile.
Dutch accounting software such as Moneybird or Exact appears in neither vendor's documents. I do not know if it works. If your administration runs there, test that first on a small task.
Does it fit a small business?
It depends on who works with it. A one-person business or a small team without a ChatGPT Business workspace has one realistic route today: Grok Bot on a trial, or through a Cursor or SuperGrok plan. A company that already pays for ChatGPT Business reaches Dots quickly through Business Premium, though the reported minimum of two seats is a real step.
Starting takes little time with either. Creating a dot takes about five minutes, according to The Neuron. With Grok Bot you write down a role, watch once and correct, and then set up a routine with a schedule. The real work is the same for both: writing down what the agent may do and when it has to ask. Both vendors mention invoices. OpenAI describes one sent after approval, SpaceXAI describes invoice processing from Gmail. For quotes I found no source.
Is it good for marketing and research?
The evidence is thin. Dots scored well in one hands-on test for marketing work: a content plan at 9.8 out of 10 and a branded website at 9.6, using a custom skill. The same reviewer gave a weekly report 8.6, and only after an explicit request for a PDF. With Grok Bot, one reviewer's weekly routine for answer-engine visibility ran reliably every Friday. A writer with an SEO Bot and Semrush also worked, though design came out weaker than with Claude. A lead-scouting Bot delivered nothing in week one.
No independent benchmarks exist, so I cannot rank them. Both browse the web, read pages, write text and fill in forms. The differences that count are in the table: where you can start, which route to access you have, and how well it works with your tools. For research one more question counts, and it comes back in a moment: does the agent name its sources, and does it say what it could not verify?
My bots and my research
Here are my two own experiences. They involve different tools. I use Grok Bot for leads. I had Claude do our own marketing research. I have not used Dots.
How I use Grok Bot
I let the bots search for specific leads, very specific requirements. I work with four bots that pass the ball to each other. The scout searches. When there is a match, the stalker picks it up and finds out all it can about the company. Then the analyst estimates the chance that it becomes something. Finally the head of desk gives me all the filtered prospects that are worth approaching.
That gives me a pre-selection. I only see what got past the other three, and I decide whom to approach. I have no figures: how many leads, how much time it saves, what it delivers. Take it as an example of what a chain of bots can do. I have no proof that it pays off.
14 October: four investigations in one day
On 14 October 2026 I wanted to know how well people find Make it Matter in Veenendaal and the region. I had Claude (Anthropic) work it out, through Claude Code, with several agents at once. Grok Bot and Dots played no part.
I split the question into four investigations and ran them side by side: an audit of our own site, what works for local findability in 2026, the five strongest competitors, and more than 60 agencies in the region. That gave five documents of about 13,500 words in total, and one plan that combined them.
On none of the 12 local search terms we checked did we find Make it Matter. Only on our own name. That is an indication, and the limits further down explain why. The biggest cause was ours:
- No page about Veenendaal or the region, and address and phone number only on the home page.
- Our second location mentioned nowhere. The town name was in the title of 4 pages and in no heading, and 12 of our 13 blogs left Veenendaal out of the title.
- No FAQ markup, no link to our Google Business Profile and an outdated sitemap template.
Outside the site it was messy too. Two phone numbers, three addresses, an old domain still visible, varying founding years, and "B.V." on the site while the trade register lists a sole proprietorship. The key figures on our own home page showed zero.
INOMA in Ede (110+ Google reviews by its own claim, 130+ articles, pages per town) and MarketingConcurrent came out as the strongest competitors. The research also warned against ten town pages with identical text. Google treats those as doorway pages, which count as spam. A page per service with the town in it, plus one service-area page with real cases, is the better route.
On 14 and 15 October we got to work. We corrected the legal form in the data for search engines (sole proprietorship, with chamber of commerce number and founding year 2010). Address, phone and email now sit at the bottom of every subpage, and llms.txt, the file for AI assistants, got facts. The link to our Google Business Profile is in the data for Google. That profile shows 4.9 stars from 15 reviews. We also built a system that posts every new blog to the profile, and published English versions of our 13 blogs.
We have no ranking results yet. It is too early to measure position effects, and I will not claim any until there are figures. Our Google Business Profile checklist and our piece on getting visible in ChatGPT and AI search cover the measures.
The method had limits:
| What I did | What it delivered | Limit |
|---|---|---|
| Four investigations side by side | In one day what a person does in several days, and one plan | Clear boundaries per agent were needed |
| Check of 12 local search terms | Indication that we do not appear locally | The agents saw no real search results with a Veenendaal location, no Google Maps and no AI overviews |
| Study of local SEO in 2026 | Concrete measures | Much of the source material is American, so applying it to the Netherlands is partly an assumption |
| Competitor profiles | A clear ranking of the strongest | Team and review data of competitors could not be verified |
| Questions back to me | Legal form, chamber of commerce number, main phone number, addresses | Only I know those facts, so human fact-checking stayed necessary |
I check our starting position myself in an incognito window. The agents delivered fast and wide. The accuracy came from a person with a primary source.
What I demand from any agent
After both experiences I ask the same of every agent, and that includes Dots and Grok Bot.
It names sources with every claim. I need to be able to click through to the page it read. It reports honestly what it could not verify. Every report in the research contained such a section, and it made the plan more reliable. Actions with consequences go past me: sending, paying, publishing and deleting I do myself.
I check prices and availability against one primary source. During this blog project the Grok Bot price the press quoted at launch turned out to be outdated within two months, and Dots availability differs by region. An agent has to ignore hidden instructions in web pages and report them. During the research for this blog, one page contained text that posed as a system instruction. The agent ignored it and reported it. An agent that follows such text can do damage in your accounts.
How to choose
A few situations you may recognise:
- You are a freelancer or work with a small team, with no ChatGPT Business workspace. Start with Grok Bot. Dots Pro is closed to you.
- You look for leads and want to keep only the best. I do that with Grok Bot: four bots in a chain that ends in a list worth your time. Start with one narrow set of requirements.
- You already pay for ChatGPT Business. Ask OpenAI about Business Premium and try Dots on one task. Dots is also the alternative once it reaches the EEA, though OpenAI gives no date.
- You work in many different apps. Check the Dots plugin list first. A tool without an API or plugin you can hand to Grok Bot, which operates the interface, though testers call that fragile.
- You handle personal data of customers. Wait with both and have a lawyer review the data processing agreement, the US location (Grok Bot) and the missing EU residency (Dots).
What to avoid: put no customer data in an agent in the first month, link no payment method, and give no access to your bank. Begin with a task you can check afterwards, such as a competitor summary with links, and compare the result with what you would have found yourself. Set the agent to ask before it sends, publishes or pays.
Frequently asked questions
What is the difference between Dots and Grok Bot?
Dots is OpenAI's always-on agent inside ChatGPT with its own cloud computer per dot, announced on 29 September 2026. Grok Bot is SpaceXAI's agent on Cursor infrastructure, with one shared computer per user, announced on 11 August 2026.
Which one is free?
Neither has a permanent free tier. Grok Bot has a one-off trial credit for accounts on a free Cursor plan, with a 7-day window and an unpublished size. I found no free trial for Dots.
Which one is available in the Netherlands?
Dots is available through Business Premium and an enterprise beta, but the personal Pro licence excludes the EEA. Grok Bot has no official region list, and I use it from the Netherlands through Cursor Pro+.
Which one is safer?
Dots gives each dot its own computer, and Grok Bot shares one computer between all Bots of a user. Both vendors admit that prompt injection is not solved, so keep approval on for every action with consequences.
Can they read my email?
Yes, if you link your mailbox. Dots can connect personal email and Gmail or Outlook through plugins, and Grok Bot has connectors for Gmail and Outlook. In both, a draft is not permission to send, but a Dots reviewer reported an email that went out unnoticed.
May I use them under the GDPR?
I found no statement about GDPR conformity from either vendor. Grok Bot computers run in the US and Dots has no data residency in the enterprise beta, so you need a data processing agreement and a legal review before you use customer data.
Do you use Dots or Grok Bot yourself?
I use Grok Bot from the Netherlands for lead research, through Cursor Pro+. I have not used Dots, and I had Claude agents do our own marketing research.
Want this kind of research done for your own company, covering online marketing and how findable you are in Google and in AI assistants? Plan an introduction at makeitmatter.studio/en#kennismaken.
Sources
- OpenAI, Introducing Dots, 29 September 2026
- OpenAI Help Center, Getting started with your dot, read on 15 October 2026
- OpenAI, DevDay 2026 recap, read on 15 October 2026
- TechCrunch, OpenAI launches Dots, 29 September 2026
- NBC News, OpenAI launches Dots, AI agents and safety questions, read on 15 October 2026
- The Neuron, OpenAI Dots hands-on review, read on 15 October 2026
- AI Agents Library, OpenAI Dots review, read on 15 October 2026
- Beri, OpenAI Dots enterprise beta, data residency and audit gaps, read on 15 October 2026
- SpaceXAI, Introducing Grok Bot, 11 August 2026
- SpaceXAI, Grok Bot in more plans, 26 August 2026
- SpaceXAI, The X Subscription, 12 October 2026
- SpaceXAI docs, Grok Bot security, read on 15 October 2026
- Cursor Help, Grok Bot plans and trial, read on 15 October 2026
- Cursor pricing, read on 15 October 2026
- The Next Web, Grok Bot, Claude Opus, Midjourney and Suno, read on 15 October 2026
- Gizmodo, Grok Bot is now a model router, read on 15 October 2026
- Sudheer Paturi, One week with Grok Bot, read on 15 October 2026
