Grok Described Our Business. It Was Wrong.
Grok told us Pixelmojo is an AI image generation tool. We are an AI visibility platform. That answer came from our own product run on July 9, 2026, the day we switched on Grok coverage in Radar and pointed it at our own brand first.
Nobody at xAI decided that. No competitor planted it. Grok simply filled a gap in its evidence with a confident guess, the way large language models do. Anyone who asked Grok about us got the wrong business, and until we measured it, we had no idea.
That is the whole argument of this post in one story. A fifth AI answer engine now describes your brand to a large audience, you probably have not checked what it says, and checking is now cheap. We will show you our measured data, explain how we added Grok to Radar without touching anyone's score, and give you a checklist you can run today.
Why Does Grok Suddenly Matter for Brand Visibility?
Grok matters because it sits inside X, where product conversations, complaints, and recommendations already happen, and because its usage is now large, audited, and public. SpaceX disclosed approximately 117 million monthly active Grok users as of March 2026, out of roughly 550 million X monthly active users, in its S-1 filing with the SEC (May 2026, as reported by Forbes). That makes Grok usage one of the few AI engine numbers that is not a vendor estimate: it is an audited securities disclosure.
Grok is xAI's assistant, available on X and at grok.com. The distribution detail is the strategic one: a user reading a thread about tools in your category can ask Grok about you without leaving the app. The question happens at the exact moment of evaluation.
Our take: for most brands the risk is not that Grok is hostile. It is that Grok is unsupervised. ChatGPT visibility gets tracked because everyone knows to track it. Grok is where the same class of wrong answer lives without anyone watching.
What Does Grok Actually Say About Brands? Our Measured Data
The honest answer: it depends on your evidence footprint, and the only way to know is to measure. Here is what we measured about our own brand on July 9, 2026, through the Grok API (grok-4.3), the day Grok coverage went live in Radar.
Our own brand through Grok, measured July 9, 2026
Production Radar audit of pixelmojo.io via the grok-4.3 API
of citation-check queries mentioned Pixelmojo
source URLs returned by the API path we measure
confidently wrong business description of our brand
Three observations from that run:
First, the misidentification. Asked what Pixelmojo is, Grok answered that it is an AI powered creative tool, likely for image generation or pixel art enhancement, and hedged that exact details are not documented. Wrong business, delivered fluently. This is the same entity confusion problem we have documented across other engines, now on a fifth surface.
Second, the mention rate. Grok mentioned Pixelmojo in 25% of the brand and category queries in our citation check. That number is one brand on one day, not a benchmark, but it gave us something we did not have before: a baseline to move.
Third, the shape of the answers. The API path we measure returned long, detailed answers and no source URLs. Which leads to the next point.
“Grok described Pixelmojo as an AI image generation tool. Wrong business entirely. Everyone who asked got a confident answer, and until we measured it, we had no idea.”
Radar production run, July 9, 2026
How Is Grok Visibility Different From ChatGPT Visibility?
Grok visibility is a mention game, not a citation game. On the API path Radar measures (verified July 9, 2026), Grok returns answers without source URLs. Perplexity cites sources in its responses; Grok, on this path, names and describes brands without linking to them.
That changes what you optimize for:
| Dimension | Perplexity-style engines | Grok (API path we measure) |
|---|---|---|
| Primary signal | URL citations with links | Brand mentions and descriptions |
| What failure looks like | Competitor page cited instead of yours | Wrong description, or absence from category answers |
| What to measure | Citation rate per query | Mention rate and description accuracy |
| Traffic mechanics | Click-through from cited link | Brand impression, no link to click |
A mention-only engine raises the stakes on accuracy. When there is no link for a user to verify, the description IS the experience of your brand. If Grok says you are an image generation tool and you sell visibility software, there is no blue link underneath to rescue you.
One measurement honesty note, and it applies to every engine we cover: Radar measures provider APIs. Consumer apps like the Grok experience inside X can layer memory, personalization, and live search on top, so answers there may differ. We publish that caveat on our methodology page because a number you cannot scope is a number you cannot defend.
The Fifth Engine Problem: Why Adding Engines Quietly Breaks Scores
Here is the part most visibility tools will not tell you: every time a vendor adds an engine to a scored average, every score changes, including the historical ones you already reported to your boss or your client.
The arithmetic is not subtle. If your visibility score averages results across four engines and the vendor adds a fifth, the denominator changes. A brand that was strong on the original four and absent on the new one drops overnight, with no change to its website, content, or actual visibility. Trend lines break silently. Our take: that is a methodology change wearing the costume of a product update.
“A score that moves when the vendor changes the formula is not a score you can defend. It is a number that happened to you.”
Pixelmojo Radar methodology position
This matters more as the engine list grows. Grok is the fifth engine we measure. There will be a sixth. Any tool in this category will face the same decision repeatedly, and there are only two honest options: version the methodology loudly, or keep new engines out of the score until promotion is announced. We chose the second.
What Is Report-Only Coverage? The Framework We Shipped
Report-only coverage means the engine is measured and shown, but excluded from score math. That one sentence is the entire framework, and we shipped it as the mechanism for adding Grok.
Concretely, in every Radar report since July 9, 2026:
- Grok appears in your per-engine results: whether it mentions you, what it says, how that compares to the other engines
- Grok is excluded from your overall score, grade, share of voice, completeness, and recommendations
- The four scored engines (ChatGPT, Claude, Gemini, Perplexity) compute your score exactly as before
- Your historical scores remain comparable, because the denominator never changed
How a new engine earns its way into a score
The report-only promotion path we use for Grok
Measure
Engine runs in every audit, data collected
Report-only
Visible in reports, excluded from score math
Validate
Stability and cost observed across real audits
Versioned promotion
Announced methodology change, scores re-baselined
This is a product claim you can verify, not a promise: the score exclusion is enforced by automated tests in our codebase, and the methodology page discloses Grok as a report-only engine. If we ever promote Grok into the scored set, it will arrive as an explicit, versioned change.
We believe this should be table stakes for the category. When you evaluate any AI visibility tool, ask one question: when you added your last engine, what happened to my historical scores? If the answer is a shrug, your trend data is decorative.
How Do You Check What Grok Says About Your Brand?
You can get a useful read on your Grok visibility in under an hour manually, or in about a minute with Radar. Both paths below.
The manual method (free, ~45 minutes):
- Write down the five questions a buyer would ask in your category. Not your brand name: the category questions ("best tools for X", "who does Y for Z companies")
- Ask Grok each question (on X or grok.com) and record whether you appear, who appears instead, and how each brand is described
- Ask Grok directly: "What is [your brand]?" and grade the answer for accuracy: right business, right products, right positioning
- Repeat the same prompts on ChatGPT, Claude, Gemini, and Perplexity so you can see whether a Grok problem is a Grok problem or an evidence problem everywhere
- Log everything with dates. Mention rates only mean something as a trend
The measured method (Radar):
A paid Radar audit runs your domain through the four scored engines plus Grok as report-only coverage, and returns mention rates, competitor substitutions, description accuracy, and hallucination flags in one report. That is the citation tracking workflow we already published for the four scored engines, now with a fifth column.
What checking gets you
Our own brand, before and after we could see Grok
- No idea Grok described us as an image generation tool
- No baseline mention rate on a 117M user engine
- Buyers could act on a wrong answer, invisibly
- Misidentification caught on day one, fix list generated
- 25% mention rate recorded as the number to move
- Grok column now tracked in every audit, score untouched
What Do You Fix When Grok Gets You Wrong?
You fix the evidence layer, not the engine. Grok reads the same public footprint every other model trains and retrieves on, which is why entity fixes compound across engines instead of being engine-specific hacks.
Priority order:
- State what you are, in machine-readable form. Organization JSON-LD with a precise description, on every key page. If your schema says nothing, models guess
- Publish one canonical description. A plain statement of what your company does on your About page, with every profile kept consistent with it. llms.txt is optional: Google says Search ignores it, and no major engine documents reading it
- Kill entity ambiguity. If you share a name with another company or product, disambiguate explicitly. The full disambiguation playbook covers this failure mode in depth
- Align third-party descriptions. Directory listings, review sites, and press that describe you inconsistently feed inconsistent answers. The GEO playbook covers building that citable footprint
- Re-measure on a schedule. Model updates change answers. A description fixed in July can regress in October, and only a dated log tells you
We are running this loop on ourselves. The Grok misidentification from July 9 went into our own fix queue. As of October 2026 we have not published a follow-up result; when we do, we will date it here.
The Bigger Pattern: Engines Multiply, Your Evidence Layer Does Not
The strategic read, labeled as ours: brand visibility work is consolidating into one evidence layer serving many engines. Grok is engine five for us. The next one will not require a new playbook, because the playbook was never really about ChatGPT or Grok. It is about whether the public record of your company is clear, consistent, and machine-readable enough that any model describing you gets you right.
That is also why we built the measurement side the way we did. Engines will keep arriving. If every arrival silently rewrites your score, you do not have a measurement program, you have a slot machine. Report-only coverage is our answer: see the new engine immediately, trust your trend line permanently.
Grok Brand Visibility: Questions Readers Ask
Common questions about this topic, answered.
What does Grok say about brands?
Grok answers brand and category questions the same way ChatGPT or Gemini do: it names companies, describes what they do, and recommends options. Whether it gets your brand right is an open question until you test it. When we ran our own brand through the Grok API (grok-4.3, July 2026), it described Pixelmojo as an AI image generation tool, which is the wrong business entirely.
How do I check what Grok says about my brand?
Two ways. Manual: ask Grok on X or grok.com the questions your buyers would ask (who does X, best tools for Y, tell me about your brand) and record what comes back. Automated: run a paid Radar audit, which can query the Grok API as report-only coverage alongside ChatGPT, Claude, Gemini, and Perplexity and shows Grok mentions, competitor substitutions, and description accuracy in one report.
Does Grok affect my Radar score?
No. Grok is report-only coverage in Radar. It appears in your per-engine results so you can see how Grok describes you, but it is excluded from every score, grade, and completeness calculation. Your score is computed from the same four scored engines as before, so historical comparisons stay valid.
Why does Radar not score Grok yet?
Because adding an engine to a scored denominator silently changes every score, including historical ones. A score that moves when the vendor changes the formula is not a score you can defend. We publish Grok data as report-only first, and promotion into the scored set will be an explicit, versioned methodology change, not a quiet update.
Is Grok visibility different from ChatGPT visibility?
Yes, in two ways. First, the audience context: Grok is embedded where conversations about products and companies are already happening on X. Second, the retrieval behavior we observe: the Grok API path we measure returns answers without source URLs, so brands are mentioned rather than cited with links. Mention accuracy matters more there than link placement.
What is report-only coverage in AI visibility measurement?
Report-only coverage means an engine is measured and shown in results but excluded from score math. It is how a measurement product adds a new engine without breaking score comparability: users see the new data immediately, and the scoring methodology only changes when the change is announced and versioned.
What should I fix if Grok describes my brand incorrectly?
The same evidence layer that fixes other engines: a clear entity definition on your site, consistent Organization JSON-LD and consistent third-party descriptions. Grok reads the same public evidence other models train and retrieve on, so entity fixes tend to help across engines.
How many people use Grok in 2026?
SpaceX disclosed approximately 117 million monthly active Grok users as of March 2026, out of roughly 550 million X monthly active users, in its S-1 filing with the SEC (May 2026). It is the first audited disclosure of Grok usage, and it is large enough that what Grok says about your brand reaches real buyers.
Does Radar measure the Grok app on X or the Grok API?
The API (grok-4.3 as of July 2026). Radar measures provider APIs for every engine, which gives repeatable, comparable snapshots. Consumer apps can layer memory, personalization, and live search on top, so answers there may differ. That is disclosed on every Radar report.
Find Out What Grok Says About You
We found out what Grok says about us by accident of process: we test our own product on ourselves first. The answer was wrong, and now it is a tracked, dated item in a fix queue instead of an invisible leak.
Your turn takes about a minute. Run your domain, read the Grok column, and see whether the fifth engine knows who you are.
Ready to see your brand through five AI engines?
- Run a free Radar check - Your technical baseline; paid audits add the AI answer checks, with Grok as report-only coverage
- Read the methodology - Exactly what we measure and why the score is defensible
- Contact Us - Talk through your AI visibility strategy
