ChatGPT vs Gemini vs Copilot vs Perplexity vs DeepSeek (2026)
We compared ChatGPT, Gemini, Copilot, Perplexity and DeepSeek again, as they are today. How to read the DISC personality comparison from our earlier article, what each AI is good for now, how to set the tone yourself, and what to check before using them at work.

HowtoAI · AI chatbots compared · Latest
?Who this is for
- You are not sure how ChatGPT, Gemini, Copilot and the others differ
- You heard that AIs have personalities and want to use that to choose one
- You want to narrow the AI tools you use at work down to one or two
!What you will learn
- The models behind all five as of October 2026, and what changed
- How to read the DISC results from our 2025 article today
- How to set an AI's tone yourself, and what to check before using one at work

Hi, I'm Hikari.
In March 2025 we published an article in Japanese called "Do AIs have personalities?", comparing five AI chatbots by personality type. In this article I'll first check where those five stand now, then look back at how we compared them. After that I'll cover what each one is like today, how to set the tone yourself, and what to watch for before you use them at work.
Where the five AIs are now
Our 2025 article compared ChatGPT, Copilot, Perplexity, DeepSeek and Gemini. As of October 2026, all five are still running. But the AI models inside them, and the way you use them, have changed a lot since then.
Here are the main changes of the past year or so, service by service.
| Service | Main model or form, October 2026 | Biggest changes this year |
|---|---|---|
| ChatGPT (OpenAI) | GPT-6. Paid Plus, Pro, Business and Enterprise get GPT-6 Sol; Free and Go get GPT-6 Luna. | GPT-6 began rolling out on October 7, 2026. It came with "Intelligent UI", which can answer with diagrams, buttons, forms and charts depending on the question. |
| Gemini (Google) | Gemini 3.5 Flash and 3.6 Flash for everyone. Gemini 4 Argon on paid Google AI Pro and up. | Gemini 3 Pro in November 2025, 3.5 Flash in May 2026, 3.6 Flash in July. On October 13, Gemini 4 Argon started rolling out to AI Pro and Ultra subscribers. "Gems", which store your instructions, are being replaced by "skills". |
| Microsoft Copilot | The consumer Copilot app and the "Microsoft 365 Copilot" app became a single "Microsoft Copilot" app. | The merged app rolled out from August 2026. Podcasts, group chat and Deep Research in the consumer app ended on August 18. |
| Perplexity | A service that lets you switch between AI models from several companies, including Anthropic, OpenAI and Google. | In February 2026 it announced "Model Council", which asks several models at once and combines their answers, and "Perplexity Computer", an agent that carries out tasks for you. Both started on the top Max plan. |
| DeepSeek (China) | The DeepSeek-V4 series. The web and app models keep being updated. | V4 released in April 2026. V4-Pro came to the web and app in August. V4.1-Flash followed in September, and an update of it on October 17. |
ChatGPT, Gemini and DeepSeek all released a new generation of models in 2026. Copilot changed the shape of its app, and Perplexity changed how it combines models. The "five AIs" we compared in March 2025 have the same names, but it is more accurate to treat them as different products now.
HIKARI'S TAKEAll five are still around, but what's inside has changed a lot. Please don't assume the "personalities" measured in 2025 still fit them today.
Checking our 2025 article
Our March 2025 article compared the five AI chatbots using DISC, a model that sorts human behavior into four styles: D for Dominance, I for Influence, S for Steadiness and C for Conscientiousness.
It was based on an article by Online DISC Profile, a company that sells DISC assessments. That article placed the five AIs in these types.
| AI | Type then | How our 2025 article described it |
|---|---|---|
| ChatGPT, Copilot | DI | Confident and quick to decide. Suited to sales and decision-making. |
| Perplexity | ID | Outgoing and persuasive. Suited to social media and marketing. |
| DeepSeek | SIc | Calm and empathetic. Suited to internal documents and HR. |
| Gemini | SCi | Careful and precise. Suited to FAQs and customer support. |
We read the original article again. It does not say when the assessment was done, on which models, or by what method. We also cannot tell how many times it was run. The original article itself advises against using AI output as is and says to always edit it. With that in mind, here is how what we wrote then compares with today.
| What we wrote | Looking back | Verdict |
|---|---|---|
| ChatGPT and Copilot are DI, Perplexity is ID, DeepSeek is SIc, Gemini is SCi | We reported the original article's results correctly. But the original never stated its method or timing, and the models it tested have since been replaced. | Half right |
| ChatGPT and Copilot have the same personality | Microsoft does offer OpenAI's new models in its business Copilot. But Copilot has become an app that combines personal and work accounts, and its screens and settings are separate from ChatGPT's. | Half right |
| Perplexity is outgoing and persuasive, good for social media and marketing | Perplexity moved to a design where you switch between models from several companies. No single personality describes it. | Not anymore |
| Gemini or DeepSeek is recommended for HR and training teams | We gave no basis for choosing by department. We also left out DeepSeek's privacy issues, which were already public in February 2025: Japan's privacy regulator had issued a notice, and Texas, New York and Virginia had banned it on state devices. | Wrong |
| With good prompts, you can adjust an AI's "personality" to some degree | Some services now let you choose a tone in their settings. The room for adjustment has actually grown. | Right |
There is one more thing we did not see at the time: an AI's tone changes when its maker updates it. In April 2025, right after updating GPT-4o in ChatGPT, OpenAI rolled the update back because responses had become overly agreeable with users. OpenAI said the cause was giving too much weight to short-term user feedback. It also said it was testing ways to let users choose from several "personalities".
Our biggest miss was treating the assessment as a reason to choose an AI without checking where it came from or how it was done.
HIKARI'S TAKEWe reported the source correctly. But it was a one-off result with an unknown method, and an AI's tone can change with any update. Looking back, it was too weak to choose an AI by.
What each of the five AI chatbots is like in 2026
Now we compare the five as they are today. Instead of personality types, we use what each company has announced. In the table, the middle column is what official announcements and press reports confirm, and the right column is how we think each one could be used. Please read the two separately.
| Service | What announcements and reports confirm | Where it may fit (our view) |
|---|---|---|
| ChatGPT | Answers with tables, diagrams and interactive parts depending on the question. You can pick a base style from eight options. Free users get GPT-6 Luna. | Sorting out your thinking, first drafts of explanations, and a first stop when you don't know what to ask. |
| Gemini | With Google AI Pro, you can use Gemini inside Gmail and Google Docs. Google pitches Gemini 4 Argon at long, complex tasks and knowledge work such as law and finance. | Everyday email and document work at offices that run on Google's tools. Reading long documents. |
| Copilot | One app switches between a personal account and a work or school account. The two sets of data stay separate. | Offices where Word, Excel and Outlook are the center of the work. |
| Perplexity | Lets you switch between models from several companies. Model Council, on higher plans, asks three models at once and shows where they agree and where they differ. | Groundwork for research, and checking an important answer against another AI's answer. |
| DeepSeek | Its web and app models keep moving through the V4 line. Its privacy policy says the data it collects is stored on servers in China. In October 2026 the Department of Homeland Security ordered federal civilian agencies to remove DeepSeek products. | Trying it out or comparing it with other AIs, without entering personal or confidential information. |
What the table shows is that what separates the five today is less their "personality" and more where they run and what they connect to. Gemini is built into Google's services, and Copilot into Microsoft 365. Perplexity's defining feature is that it doesn't depend on any one model. In February 2026, a Perplexity executive told TechCrunch that "multi-model is the future."
Seen that way, it is more practical today to choose by which tools you already work in and what information you are allowed to give an AI, rather than by which AI seems outgoing.
HIKARI'S TAKEThe real differences now show up in which work tools each AI connects to, more than in personality. Start from the email and document tools you already use, and the choice gets easier.
You can set an AI's personality: ChatGPT and Gemini
Our 2025 article said you could change an AI's personality to some degree with prompts like "please be gentler." Things have gone further since: some services now let you choose a tone in the settings.
ChatGPT added eight base styles in November 2025, with the release of GPT-5.1: Default, Professional, Friendly, Candid, Quirky, Efficient, Nerdy and Cynical. You choose one in settings and can change or undo it at any time. In December 2025, personalization settings also gained controls for warmth, enthusiasm, the amount of headers and lists, and the amount of emoji, each set to more, default or less.
Gemini announced "skills" on September 30, 2026. A skill saves instructions you use often, and you call it by typing "/" in the prompt box. Existing Gems will start moving to skills automatically on November 17, 2026. On October 13, Gemini also added a "Customize" menu for setting how it responds.
For Copilot, Perplexity and DeepSeek, we found no similar announcement of a feature for choosing tone. With these, putting your instructions at the start of the conversation still works.
If you set a work tone in this order, you won't get lost.
- Write your rules for outside-facing text in three lines. For example: "Use a professional tone," "Lead with the conclusion," "No emoji."
- Save them in AIs that have settings. In ChatGPT, use the base style and fine-tuning controls. In Gemini, use Customize or a skill.
- Paste them at the start of the chat in AIs that don't. Keep the text in a note so you can reuse it every time.
- Give the same request to two AIs and compare. Whichever needed less editing gets that job.
HIKARI'S TAKEAn AI's tone has gone from something you pick to something you set. If you write your company's style rules in three lines, you can bring any AI close to the same voice.
Plans for the five AI chatbots, and what to check before using them at work
All five can be started for free. Paid plans change which models you get and how much you can use them. Prices change often, so check the official pages before you sign up.
Paid plans are Go, Plus and Pro. Go is $8 a month and Plus is $20. Pro comes in three tiers: $100, $200 and $500 a month. Plus and Pro get GPT-6 Sol; Go gets GPT-6 Luna, like Free.
Free accounts get the Flash models. Gemini 4 Argon needs AI Pro or higher. AI Plus dropped to $4.99 a month in June 2026.
The free tier covers chat, image creation and file uploads. Higher limits and agent features need a Microsoft 365 subscription.
Model Council and Perplexity Computer started on the top Max plan. Check the official page for what the free tier and Pro include.
The web and app run V4-series models. The developer API is billed by usage.
US prices for Google AI Plus and Pro are from Google's US plans page (checked October 26, 2026); Engadget (June 2026) reported the AI Plus price cut. ChatGPT prices are from Android Authority (January 2026) and OpenAI's help center (Pro tiers, checked October 26, 2026). Perplexity Max is from TechCrunch (February 2026).
More important than price are the four points below. They apply to all five if you use AI-generated text or images in your work.
- Check rights. Under general U.S. copyright principles, output that copies protected expression from an existing work can infringe, no matter what tool produced it. The U.S. Copyright Office also concluded in January 2025 that AI output is protected only where a human has determined sufficient expressive elements, and that prompts alone are not enough. Before publishing text or images, check that they don't resemble existing work, and don't assume you own them outright.
- Don't publish it as is. Have a person check facts, numbers and names against the original sources.
- Decide what goes in. Before entering customer data or confidential information, find out where the service stores data and how it handles it. DeepSeek's privacy policy says it stores the data it collects on servers in China.
- Keep what you make. In the consumer Copilot app, group chat content was wiped in August 2026 and podcasts could no longer be viewed. Save important conversations and outputs to your company drive.
HIKARI'S TAKEYou can try every one of these for free. Before you think about price, decide what information may go in and what gets checked before it goes out. And don't leave what you make sitting inside the service.
What to decide before personality: choosing an AI chatbot for a small company
Over the past year, all five AI chatbots changed a lot inside, and Copilot even changed the shape of its app. Six months from now they may look different again. So we have stopped choosing AIs by personality, and decide in this order instead.
- Start from the tools you already use. If your work centers on Gmail and Google Docs, try Gemini. If it centers on Word and Outlook, try Copilot. If neither, start with ChatGPT.
- Pick one main AI and one for checking. Ask a second AI about important answers and compare. Where they disagree is where a person needs to look.
- Write down what information may go in, first. Decide how to handle personal data, client information and unreleased figures before you start.
- Set the tone in settings. Put your company's style rules into the settings or a standing instruction.
- Review every six months. Check models, prices and discontinued features, and update how you use them.
Our 2025 article was useful as a prompt to notice that AIs differ. But we now think an AI's character isn't something you measure once. It's something you shape to fit how your company works.
HIKARI'S TAKEAIs change quickly, so a fixed order for choosing lasts longer than memorizing personalities. Start today with one AI that sits close to the tools you already use.
Common questions about comparing AI chatbots and their "personalities"
Do AI chatbots really have personalities?
Not in the human sense; they have no personality or consciousness. They do have tendencies in how they talk, but those change with model updates and user settings. In April 2025, a ChatGPT update made responses overly agreeable, and OpenAI rolled it back.
Are the DISC types from the 2025 article still useful?
We don't recommend applying them as they are. The source article didn't state its method or timing, and all five AIs have changed a great deal since.
Is it OK to use DeepSeek for work?
DeepSeek's privacy policy says it stores the data it collects on servers in China. On October 20, 2026, the Department of Homeland Security published an order requiring federal civilian agencies to stop buying DeepSeek products and remove them from their systems. Texas, New York and Virginia banned it on state devices in early 2025. These rules cover government use, not private users or companies. Still, don't enter customer information or confidential data, and check your company's rules before using it.
How do I change ChatGPT's tone?
In personalization settings, you can choose a base style. You can also set warmth, emoji and similar options to more, default or less. In AI chatbots without such settings, start the conversation with an instruction such as "Use a professional tone and lead with the conclusion."
SOURCES
- Online DISC Profile, "Do AI Tools Have A Personality Type?" (no date given, checked October 26, 2026) onlinediscprofile.com
- Android Headlines, "OpenAI Is Finally Bringing GPT-6 to ChatGPT for Everyone" (October 7, 2026) androidheadlines.com
- Android Authority, "ChatGPT's best paid plan finally comes to the US for the cost of a sandwich" (January 16, 2026) androidauthority.com
- Thurrott, "OpenAI launches $100/month ChatGPT Pro plan for heavy Codex users" (April 10, 2026) thurrott.com
- TechCrunch, "OpenAI explains why ChatGPT became too sycophantic" (April 29, 2025) techcrunch.com
- SD Times, "OpenAI's latest update delivers GPT-5.1 models and capabilities to give users more control over ChatGPT's personality" (November 13, 2025) sdtimes.com
- Engadget, "You can now tweak how warm and enthusiastic ChatGPT's responses are" (December 21, 2025) engadget.com
- Google, "Gemini Apps release notes" (checked October 26, 2026) gemini.google/release-notes
- Thurrott, "Google's Gemini 4 Argon Model Comes to the Gemini App for AI Pro and Ultra Subscribers" (October 13, 2026) thurrott.com
- Engadget, "Google cuts the price of its AI Plus plan and doubles the storage" (June 8, 2026) engadget.com
- 9to5Google, "Gemini app replacing Gems with skills in November" (September 27, 2026) 9to5google.com
- The Register, "Microsoft's dueling Copilot apps have combined into a single entity" (August 13, 2026) theregister.com
- SD Times, "OpenAI's GPT-6 Astra now available in Microsoft applications" (September 8, 2026) sdtimes.com
- TechCrunch, "Perplexity's new Computer is another bet that users need many AI models" (February 27, 2026) techcrunch.com
- Perplexity Changelog, "Upgraded Deep Research, Model Council" (February 6, 2026) perplexity.ai/changelog
- DeepSeek API Docs, "Change Log" (checked October 26, 2026) api-docs.deepseek.com/updates
- DeepSeek Privacy Policy (updated February 10, 2026, checked October 26, 2026) cdn.deepseek.com
- Federal Register, "Notice of Order Regarding Exclusion and Removal of DeepSeek Products From Federal Civilian Executive Branch Information Technology Systems and Procurements" (October 20, 2026) federalregister.gov
- BleepingComputer, "DHS bans DeepSeek products across US federal civilian agencies" (October 20, 2026) bleepingcomputer.com
- Data Privacy + Cybersecurity Insider, "Three States Ban DeepSeek Use on State Devices and Networks" (February 13, 2025) dataprivacyandsecurityinsider.com
- Personal Information Protection Commission, Japan, notice on DeepSeek (in Japanese, published February 3, 2025, updated March 5) ppc.go.jp
- U.S. Copyright Office, "Copyright Office Releases Part 2 of Artificial Intelligence Report" (January 29, 2025) copyright.gov
- U.S. Copyright Office, "Copyright and Artificial Intelligence" (checked October 26, 2026) copyright.gov/ai
- OpenAI Help Center, "About ChatGPT Pro tiers" (checked October 26, 2026) help.openai.com/en/articles/9793128
- Google AI plans (United States, checked October 26, 2026) gemini.google/us/subscriptions
HowtoAI is the AI how-to series we have run in Japanese on multifverse.com since 2024. With our company now named ATATAKAI Digital, the series continues from October 2026 as one category of this blog. Thank you for reading, and we hope you will keep following along.
Which AI, for which work?
An AI's tone and strengths change with updates and settings. We help you decide which AI to use and how, based on the tools your company uses and the information it handles.
Talk to ATATAKAI Digital
Updated Oct 26, 2026
Read next

Google Whisk has shut down: our old how-to, and how to use Flow now
We rewrote our guide to Google Whisk now that the tool has ended. Whisk shut down on April 30, 2026, and its features moved to Google Flow. What happened, what our original article got right and wrong, and how to combine images in Flow now, with notes on rights.
