ScruTool

ChatGPT

The most capable consumer AI assistant, with cited live search, autonomous Deep Research and a much-improved image model. Powerful, but it states false facts with total confidence.

Arden Presley
Reviewed by
Arden Presley · Tech Reviewer
Updated On
Sep 18, 2026
Scrutool Score
7.5 /10
★★★★☆
Recommended

The most capable consumer AI assistant by feature count, with genuinely useful cited search and a much-improved image model, held back by confident answers that are sometimes flatly wrong.

Ease of use & availability 9.0
Live web search & citations 8.5
Deep Research (Plus) 8.5
Image generation 8.5
Answer accuracy & reliability 6.0
Free-tier value 6.0
Category AI Assistant Chatbot Productivity Image Generation Deep Research Web Search
Platform Type
Freemium web · Windows · macOS · iOS · Android
Free Tier
Yes US free tier carries ads
Image Generation
Strong spells text, counts objects
Underlying Models
GPT-5.6 + GPT-6 Luna, Astra, Sol Pro
Deep Research
Plus & up no free equivalent
Live Web Search
Yes clickable source cards
30-Second Verdict

Should you use ChatGPT?

Best for

Anyone who wants one assistant for everything : drafting, cited web search, image generation and, on Plus, autonomous Deep Research reports.

Skip if

Your work cannot survive a confident wrong answer and you will not fact-check. It states false figures that look verified.

Real cost

Free is usable but ad-supported in the US. Go is $8/mo , Plus $20/mo for Deep Research and GPT-6, Pro from $100.

Watch out

Free image generation runs a hidden daily quota that can start near empty, locking you out before you make a single image.

Overview

What is ChatGPT?

ChatGPT is the assistant most people picture when they hear the word AI, and it carries the widest feature set of any consumer tool in the category. The free tier runs GPT-5.6 Luna with unlimited text chat, while the paid plans layer on GPT-6 Astra, autonomous Deep Research, the Codex coding agent, projects, custom GPTs and unlimited image generation. It runs on the web, Windows, macOS and both mobile platforms, with history synced across all of them.

The thing that actually decides whether it fits your work is not capability, since it clearly has that, but trust. In testing it produced a confident, date-stamped pricing table that was wrong by nearly five-fold, with no refusal and no glitch to warn you. The US free tier now serves ads, and image generation sits behind a demand-based daily quota that can lock you out before you create anything. So the real questions are whether you will verify what it tells you, and whether the paid features earn their place in your week.

Capabilities

Key features & how they perform

Each feature rated from hands-on testing and aggregated review sentiment.

🔎
★★★★☆ 4.4

Live web search with citations

Current questions return real, clickable source cards from named outlets with publication dates, not a bare paraphrase.

🧪
★★★★☆ 4.3

Deep Research (Plus)

Runs autonomous multi-search reports in minutes, one test logging 37 citations from 557 searches in eight minutes.

🎨
★★★★☆ 4.2

Image generation

The old failures are mostly gone: it now spells requested text cleanly and counts objects correctly at small numbers.

🧠
★★★★☆ 4.1

Reasoning and multi-step work

GPT-6 Astra on Plus handles layered prompts and long context, with a 256K window on Go and 400K on Pro.

✍️
★★★★☆ 3.6

Writing and the Canvas editor

Drafts in a side-panel editor with varied sentence length, though the prose stays competent and oddly characterless.

⚠️
★★★☆☆ 3.0

Factual reliability

The weak spot: confident answers that read as verified but can be wrong, even on a live, web-connected session.

Feature ratings blended from Trustpilot, G2, Capterra, App Store & GeniusFirms review patterns + hands-on testing.

Hands-On Walkthrough

Running the same tasks across a free and a Plus account

Drafting, cited search, image generation, Deep Research, the free-tier walls, and the one result that should worry you.

Brainstorming and everyday drafting

The most common reason people open ChatGPT is to turn a blank page into a first draft. I started there, on the free account.

This is also where the free tier reminds you it wants your money, with an upgrade card dropped straight into the answer.

What I asked for

"15 Instagram content angles for a small US sustainable-coffee brand aimed at Gen Z"

What I noticed The free tier answered in a few seconds with a numbered table, each angle paired with an example caption hook, usable as is. Sitting in the middle of that answer was a black Get Plus card pushing upgraded reasoning. It showed up often enough that I stopped noticing it by day two.

Live web search

ChatGPT can pull current information from the web and cite it, and this is where the free tier earns its keep. I asked for a roundup of the week's news to see how it sourced the answer.

My prompt

"the biggest US tech stories of the week with sources"

How it went It pulled current reporting and credited real outlets such as Reuters and TechCrunch, each source shown as a preview card with a thumbnail and a publication date. The citations were genuine and clickable, which closes the old gap between asking a chatbot and checking a search engine yourself.

Image generation, on the two things it used to fail

Older image models were mocked for two specific failures: rendering readable text inside an image, and counting objects correctly. I tested both on Plus.

First, text on a poster.

The input

a coffee-shop poster reading "FRESH BREW DAILY - OPEN 7AM"

Then the counting test, where the old models reliably added or dropped an object or two.

Exact prompt

"exactly six red apples on a wooden table"

The result The poster came back clean, the lettering spelled correctly and set without the garbled characters that plagued these models a year ago. The apples request produced exactly six. Both beat my expectations; to watch these models miscount today you have to climb into the teens and twenties before they slip.

Deep Research on Plus

Deep Research is the Plus feature with no real free counterpart. Instead of one answer, it writes a research plan, asks you to approve it, then works on its own across hundreds of searches.

I set it a market question and kept a free account running in a second window to compare.

Prompt I used

"research the US oat-milk market with sources"

Worth knowing The report arrived after eight minutes, the header logging 37 citations drawn from 557 searches. It opened with an executive summary and offered export buttons for Word and PDF. Eight minutes of automated searching has no equivalent on the free tier, and for anyone who researches regularly this one feature carries the subscription.

The free-tier walls, and US ads

The free limits arrived sooner than I planned for, and not where I expected. A code-fixing prompt hit a reasonable cap. The image cap was the strange one.

Observation I requested the coffee poster on the free account and was told "You're out of images," with a reset timer for that evening, despite not having generated a single image on that account all session. Image generation runs on a separate daily allowance that OpenAI adjusts by demand, so the counter can already sit near zero when you arrive. You never see the mechanism; you hit the wall and wait. There is an added cost for US readers too, since the free tier has carried ads in the United States since early 2026, which the paid plans strip out.

The reliability problem

Here is the result that mattered most, and it happened on a paid, web-connected session where you would least expect it.

I asked a simple factual question about another product's pricing.

What I typed

"what is Prezi's pricing?"

It produced a confident table stamped "As of September 2026," quoting Plus at $19 a month and Premium at $29. Then I opened Prezi's live pricing page.

My take Plus was $4 and Premium was $7. Every figure ChatGPT gave ran several times high, and the Plus number alone missed by nearly five-fold. This is the failure mode that makes hallucination dangerous: no refusal, no obvious glitch, just a polished answer dressed in a date stamp that made it read as verified. Prices shift by region and promotion, so I am reporting what appeared at the time of testing in September 2026, both screens captured minutes apart. Verify anything factual before you build on it.

Letting it write the review

For the last test I handed the assignment back to the tool. I gave Plus a detailed brief built from these findings and asked it to draft this review.

The brief

"write this review at roughly 800 words from the findings above"

During testing It delivered inside Canvas, the side-panel editor, and the prose was stronger than the robotic copy people complain about: a real opening question, varied sentence lengths, on brief. Two tells surfaced anyway. It leaned hard on long comma-chains, a rhythm that reads as machine-written once you know it, and it claimed to have cross-checked figures against documentation, a tidy assertion of diligence I could not confirm from inside the chat. Competent and characterless, like a skilled writer with nothing at stake.
Plans & Cost

ChatGPT pricing

Figures taken from ChatGPT's official pricing page for individuals.

Plan Price What's included
Free $0 Unlimited text chat on GPT-5.6 Luna · limited messages with uploads, image generation, voice, deep research, memory and Codex · ad-supported in the US
Go $8 /mo Everything in Free, plus more messages, uploads, image creation and voice chats · longer memory · 256K context window on reasoning models
Plus Popular $20 /mo GPT-6 Astra and GPT-5.6 · unlimited image creation · Deep Research · projects, scheduled tasks and custom GPTs · expanded Codex · full ChatGPT Work access
Pro $100 /mo From $100/mo · 5x more usage · GPT-5.6 Sol Pro · unlimited faster image creation · maximum deep research and memory · 400K context window

The individual plans as shown on ChatGPT's official pricing page.

⚠ What the plans page leaves out The US free tier now carries ads, and free image generation runs on a demand-based daily quota that can already sit near zero when you sign in, so you may be blocked before making a single image. Go at $8 is a cheap bump in limits, but Deep Research, GPT-6 Astra, projects and custom GPTs only unlock at Plus ($20). Pro starts at $100 for 5x usage, and the page shows Pro as a "From" price, so heavier Pro usage can cost more. No figure here is a substitute for checking a fact the model states as current.
The Balance

Pros & cons

Specific conclusions from testing and real user reviews, not generic filler.

✅ Pros

  • Live answers cite real, clickable sources with publication dates
  • Image model now spells requested text and counts small object sets correctly
  • Deep Research runs autonomous multi-search reports in minutes
  • Widest feature set: Codex, custom GPTs, projects and scheduled tasks
  • Runs on web, Windows, macOS, iOS and Android with synced history
  • Canvas editor drafts inline with varied, readable sentence rhythm

⛔ Cons

  • States false facts confidently, sometimes dressed with a date stamp
  • Sycophantic: agrees with the user rather than challenging weak ideas
  • OpenAI's own research shows newer models hallucinate more, not less
  • Verbose by default, often needs a follow-up to reach the answer
  • Weak on breaking news even with live search connected
  • Its "new ideas" tend to recycle existing content, thin on originality

Synthesized from real reviews on Trustpilot, G2, Capterra, the App Store, Reddit & WebFX · paraphrased, not quoted

Benchmarks

ChatGPT scorecard

Rated against what a general-purpose AI assistant is actually built to do.

How we score Each dimension is rated 0 to 10 from hands-on testing combined with aggregated user-review sentiment (Trustpilot, G2, Capterra, App Store). The headline Scrutool Score is the equal-weight average of all 10 dimensions below.
Dimension Verdict Score
Answer accuracy & reliability Whether the confident answer holds up Average
6.0
Live web search & citations Current results with clickable source cards Excellent
8.5
Reasoning & problem solving Layered prompts and long context Good
8.0
Image generation Text rendering and object counting Excellent
8.5
Deep Research (Plus) Autonomous multi-search reports Excellent
8.5
Writing & drafting quality Canvas editor and prose voice Good
7.0
Memory & long context Recall across a conversation Good
7.5
Ease of use & availability Interface and reach across devices Excellent
9.0
Free-tier value What you get before paying, US ads aside Average
6.0
Value & limits transparency Hidden image quota and usage caps Average
6.0
Scrutool Score Equal-weight average of all 10 dimensions
7.5
Sentiment Analysis

What users say about ChatGPT

The themes reviewers raise most often, by share of analysed reviews.

👍 Most-mentioned praise
Fast, versatile all-in-one assistant 78%
Clean, beginner-friendly interface 64%
Live search with real, clickable citations 56%
Image generation that spells text and counts 46%
Deep Research saves hours of manual work 40%
👎 Most-mentioned pain
Confident hallucinations, must verify everything 62%
Too agreeable, will not challenge weak ideas 44%
Usage limits hit sooner than expected 40%
Verbose replies bury the actual answer 34%
Free tier feels restrictive, US ads added 30%

% = share of analysed reviews mentioning each theme (Trustpilot, G2, Capterra, App Store, Reddit & WebFX)

7.5
Final Verdict

The most capable assistant on the market, as long as you never take it at its word

ChatGPT does more, on more devices, than anything else in its class. Live search returns real cited sources, the image model has quietly fixed the text and counting failures it was mocked for, and Deep Research on Plus turns an eight-minute run into a sourced report that would take a person an afternoon. The interface stays the easiest in the category, which is why most people start here. The catch is the one thing an assistant most needs to get right. In testing it handed over a confident, date-stamped pricing table that was wrong by nearly five-fold, with nothing to signal the error, and that habit is the recurring complaint across user reviews alongside its tendency to flatter rather than push back. The free tier adds friction of its own, with US ads and an image quota that can lock you out before you begin. Use it as your default for drafting, search and research, lean on Plus if you research often, and verify every fact it states as current before you act on it.

Community

Discussion

Join the discussion and share your perspective.