Learn how to use Grok for chat, web and X research, files, and voice, with current plan details, weekly usage limits, and privacy controls.
Grok AI is worth considering when your work mixes everyday questions with information moving through the web and X. You can ask it to explain a topic, inspect a document, draft copy, or investigate a developing story. The challenge is deciding which parts of its answer are established facts and which are simply circulating claims.
This guide covers Grok as an AI Chat & Assistant, with practical advice on models, subscriptions, research, files, and privacy.
Documentation and prices checked October 9, 2026. Belliste has not conducted an original hands-on benchmark for this guide. The named expert commentary below is attributed to its published source.
Grok is the assistant developed by xAI, now presented under the SpaceXAI name on its official website. It is available through its standalone website, iOS and Android apps, and X.
The current product guide describes chat, voice, image and video creation, file analysis, and connected tools. Availability and allowances depend on your account, plan, and region.
Grok's personality is adjustable. For professional work, ask for a neutral tone, concise explanations, and explicit uncertainty. A humorous answer can be enjoyable without being suitable for a client report.
The model name and the subscription name are different things.
SpaceXAI announced Grok 4.7 on September 21, 2026, emphasizing coding and knowledge work. The announcement confirms availability through Grok Build, the API, Cursor, and other developer channels.
However, the consumer pricing page still names Grok 4.6 under SuperGrok when checked for this guide. That is a reason to inspect your live model selector, rather than assume every consumer plan includes the newest developer model.
Vendor benchmark results describe performance under particular test conditions. They do not establish that Grok will produce the best answer to your own question, nor that a consumer chat reproduces an API configuration.
| Capability | A practical use |
|---|
| Web and public X search | Track announcements and developing discussions | Original source, timestamp, and confirmation |
| General chat and writing | Explain concepts or revise a draft | Accuracy, tone, and missing context |
| File analysis | Extract information from PDFs, spreadsheets, or code | Supporting passages and calculations |
| Voice | Discuss a problem hands-free | Important names, numbers, and instructions |
| Imagine | Create images and videos | Output quality and usage cost |
| Connectors | Work with supported email, files, and calendars | Permissions and the data actually retrieved |
Sources: Grok overview and official plans. These describe capabilities, not a promise of identical access on every tier.
US prices displayed on the official pricing page, checked October 9, 2026:
| Plan | Published monthly price | Main distinction |
|---|---|---|
| Free | $0 | Limited access, including web/X search and voice |
| SuperGrok Lite | Not displayed on the page checked | Listed in the comparison; verify checkout |
| SuperGrok | $30 | Higher limits, frontier-model access, and creation features |
| SuperGrok Plus | $100 | More usage and priority access |
| SuperGrok Heavy | Not displayed on the page checked | Listed in the comparison; verify checkout |
| Business / Enterprise | Check business terms or request a quote | Team administration and organization requirements |
The absence of a displayed price does not mean a tier is free. Avoid treating third-party estimates for Lite or Heavy as a confirmed checkout price.
X subscriptions and standalone SuperGrok purchases also have different billing routes. Before paying, check which account has the subscription and what benefits its current terms include.
The official usage FAQ explains a paid-plan system rolling out from June 2026: a shared weekly allowance across Grok products replaces separate daily product limits.
Different tasks consume different amounts. A short chat and a long coding task or high-quality video are not equivalent units. Check Settings → Usage for the percentage consumed, product breakdown, and reset time.
When included usage runs out, paid features pause until reset, with separate free-tier Chat and Voice access remaining. Optional extra credits and upgrades can extend usage.
This changes how you should evaluate a subscription. If you mainly need research, monitor whether creative generation is consuming the allowance you expected to spend on chat. Do not budget around an old fixed “messages per day” claim.
Public X posts can expose an announcement, reaction, or correction quickly. They can also repeat a rumor thousands of times before anyone checks it.
For a developing story, request three separate groups:
Then open the links. Check when the post was published, whether it quotes another source, and whether later corrections change the conclusion. Ten posts repeating one claim are not ten independent sources.
The X help page explains that Grok can decide whether to search public posts and the web. Do not assume every answer ran a fresh search just because the product has that capability.
In a September 2024 WIRED article by Kate O'Flaherty, Camden Woollven, then identified as group head of AI at GRC International Group, highlighted access to current X information as a benefit for following unfolding events.
That is historical expert commentary, not a review of Grok 4.7. It helps explain the product's attraction without proving today's accuracy.
For a reader, the useful distinction is between discovering a new claim and verifying it. Grok can help with the first; the second still depends on the evidence.
Ask for a decision brief, comparison, document summary, or code change. Specify the audience, length, and what a successful result must contain.
Use web search for current information. Upload a document when the answer must come from that document. Avoid mixing outside assumptions into a file-only task.
Request official sources first, dated links, and a clear distinction between fact and inference. State the country when asking about pricing.
Check whether it answered the question you actually asked. A fluent explanation can still omit your budget, timeframe, or required feature.
Ask which conclusions lack direct evidence and what information would change the recommendation. Verify consequential numbers independently.
Edit client-facing prose, test generated code, and inspect spreadsheet formulas. Preserve the source links and assumptions alongside the final deliverable.
Task + Context + Inputs + Constraints + Evidence rules + Output format
> Research this week's announcements about three AI assistants for a small marketing team. Prioritize official release notes. Separate released features from previews, include dated source links, and do not treat X rumors as confirmed. Return a short comparison table and list what remains uncertain.
For a PDF:
> Use only the attached report. Summarize the revenue trend, identify the supporting section for each claim, and show the calculations. If a requested fact is missing, say it is not in the document.
These prompts improve reviewability; they cannot guarantee correctness.
Grok's documentation lists support for analyzing several file types and connecting tools such as email, files, and calendars. Use the smallest relevant dataset and verify that the answer reflects the intended file or account.
Build and Bot extend into coding and agent work. They deserve a different review process from ordinary chat: generated changes should be inspectable, and actions affecting external accounts should remain under your supervision.
Do not assign one universal context-window number to the whole assistant. Model limits, upload restrictions, and what the application actually places into context are separate constraints.
For standalone Grok, the consumer FAQ describes an Improve the Model control under Settings → Data on the web, or Data Controls in the app. Disabling it excludes new conversations from training; voluntarily submitted feedback may still be used.
Private Chat, where available, is excluded from training, but it is not a zero-retention promise. The FAQ describes deletion within 30 days, with stated exceptions.
For Grok on X, use the separate Privacy & Safety → Data sharing and personalization → Grok & Third-party Collaborators controls. Training and personalization have separate settings.
The Enterprise FAQ says business inputs and outputs are not used for model training. A consumer subscription should not be treated as an equivalent enterprise data arrangement.
The model catalog distinguishes API models and their pricing. Grok 4.7's launch lists starting rates of $2 per million input tokens and $6 per million output tokens, with a faster variant costing more.
Token rates alone do not determine your final cost. Check the selected model, long-context conditions, reasoning usage, tools, and current billing arrangement before estimating a workload.
The API and the consumer assistant are different access routes. Consult current documentation rather than assuming a website subscription grants unlimited programmatic use.
Evaluate Grok when current web and X discussions are useful inputs to your work. Compare alternatives on your own briefs when your priority is long-form writing, document reliability, coding, or integration with an existing workspace.
A practical trial is three tasks you can independently assess: one current-news question, one document analysis, and one writing or coding task. Record factual mistakes, correction effort, and allowance consumed.
Grok offers a broad assistant with a distinctive route into current web and X information. Its value depends on whether that access helps you complete real work and whether you can inspect the evidence behind its answers.
Start free, check the model and features available in your account, and upgrade only when an observed limit justifies the cost. Open Grok and test it on a task whose answer you can verify.
Grok is xAI's conversational assistant, now presented under the SpaceXAI name. It supports chat, search, voice, file analysis, and creative tools across supported access routes.
Yes. The official pricing page lists Free at $0 per month, with limited usage. Paid plans increase allowances and access; availability can vary by account and region.
The official page checked October 9, 2026 displays SuperGrok at $30 per month and SuperGrok Plus at $100 per month. Lite and Heavy are listed without displayed prices on that page; verify checkout.
Do not assume so. The September 21, 2026 announcement confirms Grok 4.7 for Build and developer routes, while the consumer pricing page checked for this guide still names Grok 4.6. Check your live model selector.
Yes, supported Grok workflows can search the web and public X posts. Search capability does not guarantee every answer uses fresh evidence. Inspect original links and separate confirmed facts from rumors.
Current documentation describes a shared weekly usage pool across products. Tasks consume different amounts, and Settings → Usage shows consumption and reset timing. Optional extra credits can extend use.
Yes, its product documentation lists file analysis including PDFs, spreadsheets, images, and code. Ask for supporting passages and verify calculations; upload limits and access depend on your account.
For standalone Grok, disable Improve the Model in web Settings → Data or app Data Controls. Grok on X has separate training controls under Privacy & Safety. Voluntary feedback may still be used.
