AI SEO · 12 September 2026 · 7 min read
The SpaceX data story is two posts and one missing model.
Every few weeks an AI headline tells a small business the ground has moved. This one, that SpaceX engineering data is going into Grok, rests on two posts from Elon Musk's personal account. A month after the second, xAI's own model list, release notes and news pages carry no Grok 4.7, and the two minute check that shows you that is worth more than any rewrite.
What did Elon Musk actually say about SpaceX and Grok?
Two posts, three weeks apart, both from his own account.
On 20 July 2026 he wrote that SpaceX's massive corpus of world-class engineering data, excluding material blocked by ITAR, will be added during supplemental training of the 2T run. Note the tense.
On 12 August 2026, the day Grok 4.6 reached the API, he wrote that Grok 4.7 is significantly better than 4.6 and should be ready in 3 to 4 weeks, and in the same post that initial training is complete and now we're adding a massive amount of SpaceX company data in supplemental training.
That is the entire evidential base for a story you have seen a dozen times since. It is not nothing: the man who runs the company is saying what the company is doing. It is also not a vendor publication and not a model anyone can call.
Has Grok 4.7 actually shipped?
Not as far as xAI has published. Here is what its own pages carried on 12 September 2026.
- The models page calls grok-4.6 its flagship model for code and everything else, with agentic tool calling, minimal hallucinations and configurable reasoning, and the identifiers on it run to grok-4-6, grok-4.20, grok-4.20-beta and their reasoning and non-reasoning variants, with no 4.7 among them.
- That same page carries the line Last updated: August 21, 2026, nine days after the post.
- The newest model entry in the release notes is dated 12 August 2026 and reads Grok 4.6, SpaceXAI's frontier model for coding, agentic tasks, and knowledge work, is now available on the xAI API.
- The address where a launch write-up would sit, x.ai/news/grok-4-7, returned HTTP 404 when we fetched it.
Our reading of that, not xAI's: an absence on a page nobody has touched since the spring is weak evidence, and an absence on a price list edited nine days after the announcement is not.
The August estimate was three to four weeks and it is now 12 September, so the outside edge of that window has gone. Estimates slip, which is ordinary engineering, but a slipped estimate and a shipped product look identical in a headline.
Has xAI itself said anything about SpaceX data?
Nothing we found, either way. How it writes about training is what makes that interesting.
In its own announcement xAI says Grok 4.6 underwent a longer supplemental training run than Grok 4.5, with curated model-generated data for reasoning and advanced technical concepts, high-quality engineering data, and an improved optimizer and training recipe. So it describes these runs in public, at length, naming the kinds of data that went in. It names engineering data. It does not name SpaceX. It names partners readily enough: Grok 4.6 is available today in Cursor and Grok Build.
One wrinkle makes the confusion easy. The release notes call Grok 4.6 SpaceXAI's frontier model, so the company's own name already carries SpaceX inside it. A name is not a training set.
None of this is a denial. xAI has said nothing on the subject, and a gap across four pages is not a disproof. Our read is that a company publishing this much about a training run would probably mention its most quotable ingredient, so the silence is a reason to wait rather than to write the story up as settled.
Why does a barber in Kent care what version Grok is on?
Directly, not at all. The reason to read this is the reflex, not the model number.
An AI headline arrives with an unwritten instruction: something has changed, so change something. We watch small businesses obey it several times a year. They rewrite a services page around whatever was announced, buy a tool built for a feature still in preview, or pay to make a site ready for a model that has no identifier yet.
The bill is not mostly money. It is the fortnight nobody spent on what decides whether a plumber in Tonbridge or a restaurant in Sevenoaks gets named in an answer: true opening hours, a page per service and per town, reviews, and an FAQ answering what people ring up and ask.
A better model does not read a different web either. It reads the same pages faster and misunderstands them less, which raises the value of saying plainly what you do, where, what it costs and when you are open. That is the practical version, written up for this engine in Grok SEO for local business and for another in how ChatGPT picks a local business. Neither needs a line changed today.
How do you check an AI announcement in two minutes?
Every check below can be done from a phone, and none needs an account.
- Open the vendor's model list. If a model exists, developers can call it, so it has an identifier on a page like xAI's models page. Search that page for the version number in the headline.
- Open the release notes. Vendors log what shipped and when. xAI's release notes run in date order, and the newest model entry tells you where the product really is.
- Guess the announcement address. These are predictable. x.ai/news/grok-4-6 is a real page. x.ai/news/grok-4-7 is a 404. An empty slot is a hint, not a verdict.
- Look at when the page was last edited. An absence on a freshly edited page counts for far more than an absence on a forgotten one.
That is the whole method. It is the difference between reading the news and checking it.
What we are doing for clients this month
- Keeping a short list of the model identifiers each engine actually offers, and checking it before acting on any announcement.
- Putting a date and a link beside every AI claim in a monthly report, so a client can tell a shipped feature from a preview from a post.
- Changing nothing about the plan. The pages, profiles, hours and review routines we look after needed no edit on 12 August and need none today.
- Watching one thing here: whether Grok 4.7 turns up on xAI's model list. If it does, we will write up what changed for local answers.
Deliberately dull. Much of the value in this job is in the work we talk clients out of, and the visible half is in our case studies.
What you should do this week
- Take the last AI headline that made you consider changing something, find that vendor's model list, and search it for the version number.
- Work out who made the claim: a personal account, a company blog, or a news site quoting a personal account. Only one is the company speaking.
- Check when you last edited your opening hours, your prices and your service pages. If any is older than the last launch you worried about, your attention is in the wrong place.
- Ask your own trade question in Grok, ChatGPT and Google's AI Mode, in a customer's words, and note who gets named. Save the answers with the date on them.
- Add one page, or one FAQ answer, for something a customer actually asked you this month, then diary a reminder to check the model list again next month.
Three objections, answered
"He runs the company, so if he says it, it is official." He does, and it may be exactly what is happening. But an intention stated by a founder and a model you can call are different things to plan around. July said the data will be added, August said it is being added, and a month on there is still no 4.7 on the list. Believe the man and still wait for the model.
"You checked a handful of pages and called it an absence." Correct, and it is the objection we take most seriously. Nothing here shows that the training did not happen, or that 4.7 will not appear tomorrow. All we can report is what xAI has published, which on 12 September 2026 contains no Grok 4.7 anywhere we looked.
"Waiting means being last." Waiting to react is not the same as waiting to work. Preparation for these assistants is the same unglamorous handful of things, none of it dependent on a version number, so you can be ready for 4.7 without knowing whether it exists.
The verdict
Two posts on a personal account, three weeks apart, are the whole of the SpaceX and Grok story. They may well be a preview of something good. On 12 September 2026 they are also the only place it exists: xAI's flagship is still Grok 4.6, its release notes carry no 4.7, and the announcement page for it is a 404.
Treat it as an intention. Intentions are worth reading and not worth rebuilding a website for. Next time a headline says the ground has moved, spend two minutes on the model list, then go back to the hours, the pages and the reviews.
Questions, answered
Is Grok 4.7 out yet?
Not as far as xAI has published. On 12 September 2026 its models page named grok-4.6 as the flagship, its release notes carried no Grok 4.7 entry, and the address x.ai/news/grok-4-7 returned HTTP 404. Elon Musk wrote on 12 August 2026 that it should be ready in 3 to 4 weeks, and that window has now passed.
Did SpaceX engineering data really go into Grok?
Nobody outside the company can say. Elon Musk wrote on 20 July 2026 that SpaceX's massive corpus of world-class engineering data, excluding material blocked by ITAR, will be added during supplemental training of the 2T run, and on 12 August that a massive amount of SpaceX company data is being added in supplemental training. In the xAI pages we checked, nothing repeats it and nothing denies it.
Where can I check which models a vendor has actually released?
For xAI it is the models page at docs.x.ai, which names grok-4.6 as its flagship model for code and everything else and lists the identifiers a developer can call, including grok-4-6, grok-4.20 and grok-4.20-beta. It carried the line Last updated: August 21, 2026 when we read it. Every serious vendor publishes an equivalent page.
What does supplemental training mean?
In plain terms it is a further round of training on top of an existing base, rather than a new model from scratch. xAI describes one for Grok 4.6 in its own words: a longer supplemental training run than Grok 4.5, with curated model-generated data for reasoning and advanced technical concepts, high-quality engineering data, and an improved optimizer and training recipe. It names engineering data and does not name SpaceX.
Does any of this change what my business should do to its website?
No. Neither post, and none of the xAI pages we read, changes how an assistant reads a local business page. The work is what it was in July: say what you do, where, roughly what it costs and when you are open, keep the Google Business Profile accurate, and answer the questions customers ask.
Should I switch AI tools every time a new model is announced?
Rarely, and never on an announcement alone. Check the model list first, because if the version in the headline has no identifier on the vendor's own page there is nothing to switch to. When one does appear, the question is whether it changes an answer your customers see, which you can test with your own trade question.
Sources
- xAI, Models, last updated 21 August 2026
- xAI, Release Notes, entry dated 2 September 2026
- xAI, Introducing Grok 4.6, 12 August 2026
- Elon Musk on X, 12 August 2026 (Grok 4.7 and SpaceX company data)
- Elon Musk on X, 20 July 2026 (SpaceX engineering corpus and the 2T run)
- xAI, Grok 4.7 announcement address, HTTP 404 when fetched on 12 September 2026
Our own figures come from Google Analytics 4 and Google Search Console for the sites we look after, pulled 12 September 2026.


