Searching the web with AI has gone mainstream. OpenAI says search-style ChatGPT queries now equal about 20% of Google’s volume, Google has pushed AI Mode to most of the world, and Perplexity still markets itself as the answer engine built for research. All three read the web, write an answer and cite sources, but they differ a lot in price, accuracy habits, privacy and what they are best at. I compared them on the things that matter when you actually rely on the answer.
- Google AI Mode: free, fast, best for local results, shopping and very fresh info. Powered by Gemini Flash models, available in nearly 200 countries.
- ChatGPT search: best when search is part of a bigger task (compare, plan, write, code). Works on every ChatGPT plan; 1.2 billion weekly users.
- Perplexity: best for research with clear, numbered citations and source control. Free plan has 3 Pro Searches a day; Pro is $20/month, Max $200.
- Click-through: AI answers send far fewer visits to websites; Google still sends about 190x more traffic than ChatGPT.
- My pick: AI Mode for quick everyday lookups, Perplexity for research you need to cite, ChatGPT when the search feeds into a longer job.
Quick Comparison Table
| Google AI Mode | ChatGPT search | Perplexity | |
|---|---|---|---|
| Price | Free | Free to Pro; plans from $8 (Go) and $20 (Plus) | Free; Pro $20/month ($200/year); Max $200/month |
| Models | Gemini Flash family (default) | OpenAI GPT-6 family, by plan | Own Sonar models plus selectable third-party models on Pro |
| Index | Google’s own web index, Maps, Shopping | Web search partners plus OpenAI crawler | Own crawler plus web search |
| Citations | Links beside the answer | Inline source chips | Numbered inline citations, source filters |
| Best at | Local, shopping, fresh news, speed | Multi-step tasks, follow-up work | Research, academic and finance queries |
| Browser | Chrome with Gemini | ChatGPT Atlas | Comet (free) |
| API | Gemini API with Google Search grounding | OpenAI web search tool | Sonar API |
Google AI Mode
AI Mode is the AI tab inside Google Search. At Google I/O in May 2026, Google made Gemini 3.5 Flash the default model for AI Mode globally, added an “intelligent search box” that accepts text, images, files, videos and open Chrome tabs in one query, and expanded Personal Intelligence (opt-in answers using your Gmail and Google Photos) to nearly 200 countries and 98 languages at no cost. Generative UI, where Search builds custom tables, diagrams and simulations, rolled out over the summer. Google has since moved its consumer apps to newer Flash models.
Strengths: free, very fast, the best local and shopping data (Maps, reviews, prices), and fresh news. Weaknesses: answers are shorter and more commercial, and controlling which sources it uses is limited. Note that Google is tying AI tightly to Search; this week it said schools that turn off its AI tools for students must also lose Search.
Google’s AI features in Chrome are covered in my Gemini in Chrome guide.
ChatGPT Search
ChatGPT searches the web automatically when a question needs current information, or when you ask it to. OpenAI says ChatGPT reached 1.2 billion weekly users in early November 2026 and that search-like queries are about 20% of Google’s volume (OpenAI’s estimate; details in my ChatGPT search vs Google analysis).
Strengths: it keeps going after the search. You can ask it to compare the results in a table, build a plan, draft an email or write code using what it found, all in one chat. It also has memory, so it can tailor answers to you. Weaknesses: citations are less prominent than Perplexity’s, and it sometimes answers from model knowledge when you expected a fresh search; say “search the web” to force it. Free users get fewer advanced searches than paid plans, see my ChatGPT plans guide.
Perplexity
Perplexity is search-first. Every answer comes with numbered citations, and you can steer sources toward academic papers, finance data or social discussion.
| Plan | Price | What you get |
|---|---|---|
| Free | $0 | Sonar (auto model), 3 Pro Searches and 5 Deep Research per day, 40 MB uploads |
| Pro | $20/month or $200/year | Full Sonar family plus selectable third-party frontier models, Spaces with up to 50 files, $40 in Perplexity Computer credits |
| Education Pro | $10/month | Pro features with student verification |
| Max | $200/month | Perplexity Computer (orchestrates many models), Model Council, unlimited Labs and Research, highest limits |
| Enterprise Pro / Max | $40 / $325 per seat | Team controls, 400 or 4,000 Pro Searches per week |
Prices as last verified by Suprmind in September 2026. Perplexity’s Comet browser has been free worldwide since October 2025, and the $5 Comet Plus publisher add-on is included with Pro and Max.
Strengths: the clearest citations, source filters, good Deep Research reports. Weaknesses: the free plan’s 3 Pro Searches a day go quickly, and its model line-up on Pro changes frequently, so check which models are on offer.
Accuracy and Citations: What to Watch
- A citation is not proof. All three sometimes cite a page that does not say what the answer claims. Click through for anything important: prices, dosages, laws, dates.
- Fake citations are now rare in top models. A Stanford retest in November 2026 found leading models fabricate references only around 0 to 1% of the time, though some smaller models still do it often (see my Stanford shelf-life study write-up).
- Sycophancy is still real. If you phrase a question with a built-in assumption (“why is X the best…”), all three tend to agree with you. Ask neutrally.
- Freshness differs. Google is usually fastest on breaking news; ChatGPT and Perplexity can lag on very new pages.
Privacy and Personal Data
- Google AI Mode can use Gmail and Photos only if you opt in to Personal Intelligence. Search activity is tied to your Google account unless you turn off activity saving.
- ChatGPT stores chats and may use memory; you can turn off model training on your data and use temporary chats.
- Perplexity keeps history by default; the free plan lists 30-day file retention, and you can opt out of AI data usage in settings.
- For sensitive research, consider a local model with a private search tool. My best Ollama models guide is a good start.
APIs for Developers
| API | Price (per 1M tokens, in/out) | Notes |
|---|---|---|
| Perplexity sonar | $1 / $1 | Plus search context fee $5 to $12 per 1,000 requests |
| Perplexity sonar-pro | $3 / $15 | Context fee $6 to $14 per 1,000 |
| Perplexity sonar-deep-research | $2 / $8 | Plus citation, reasoning and search fees; about $0.82 per complex request |
| OpenAI web search tool | Model price plus tool fee | See my OpenAI API pricing guide |
| Gemini grounding with Google Search | Model price plus grounding fee | See my Gemini API pricing guide |
Which One for Which Job?
| Task | Best pick |
|---|---|
| “Best pizza near me open now” | Google AI Mode |
| Comparing laptops under a budget, then writing a shortlist | ChatGPT search |
| Literature review with citations for a report | Perplexity (Pro or Deep Research) |
| Breaking news in the last hour | Google AI Mode |
| Researching a stock or crypto project | Perplexity finance focus, then verify on primary filings |
| Coding question that needs current docs | ChatGPT (or Codex) with search |
| Free, unlimited everyday use | Google AI Mode |
You do not have to pick one. I use AI Mode as a default, Perplexity when I need to cite, and ChatGPT when the answer is the start of a bigger job.
How to Compare Them on Your Own Query
The best tool for you depends on what you search for, so the fairest test is your own question. This takes about ten minutes:
- Pick two or three real questions you care about and can partly check yourself: one local or shopping query, one research question and one “help me decide” task.
- Google AI Mode: sign in to your Google account, type the question on google.com and tap AI Mode under the search box, or go straight to google.com/ai.
- ChatGPT: start a new chat and begin with “Search the web for…” so it fetches fresh pages instead of answering from memory.
- Perplexity: paste the same question. On the free plan, save your 3 daily Pro Searches for the hardest question.
- Ask the same follow-up in all three, such as “Which of these is cheapest and why?”, to see how well each keeps context.
- Open two or three citations from each answer and check that the page really says what the answer claims, and how recent it is.
- Score each answer using the table below, then use the winner for that type of query.
| Criterion | What to check |
|---|---|
| Accuracy | Do the key facts match the cited pages and what you already know? |
| Freshness | Are the sources from this week or month, or older pages? |
| Citation quality | Are sources primary (official sites, papers) or thin summaries? |
| Usefulness | Did it answer the actual question, or pad it out? |
| Follow-up | Did the second question build on the first answer correctly? |
Troubleshooting: Common Problems and Fixes
| Problem | Likely cause | Fix |
|---|---|---|
| AI Mode tab does not appear | You are signed out or in Incognito, your account is a work or school (Workspace) account, you are under 18, or your language or country is not supported yet. | Sign in to a personal Google account in a normal window, try google.com/ai, and check Google’s AI Mode help page for current requirements. |
| Answer is outdated | The tool answered from older pages or model knowledge. | Add the month and year to your question, ask it to search again for recent sources, and check page dates yourself. |
| ChatGPT shows no sources | It answered from its training data without searching. | Ask “search the web and cite your sources”, or start a new chat with that wording. |
| Citation does not support the claim | The model summarised loosely or mixed up two pages. | Open the source, quote the relevant line back and ask it to correct the answer, or rely on the primary source. |
| Perplexity Pro Searches run out | The free plan includes 3 Pro Searches a day. | Use standard search for simple questions, wait for the daily reset, or upgrade to Pro if you hit the limit often. |
| Results are for the wrong location | A VPN or your account location changes where the tools think you are. | Turn off the VPN, or name the city in your question. |
| Answer is one-sided | Your question contained an assumption. | Rephrase neutrally, for example “pros and cons of X” instead of “why is X best”. |
Frequently Asked Questions
Is Perplexity better than ChatGPT for search?
For research with citations, often yes: its citations and source filters are clearer. For tasks that continue after the search, ChatGPT is better.
Is Google AI Mode free?
Yes. AI Mode is free in Google Search in nearly 200 countries. Paid Google AI plans add higher limits and newer models in the Gemini app.
Is Perplexity Pro worth $20?
If you run more than 3 Pro Searches a day or use Deep Research for work, yes. Students can get Education Pro for $10.
Which AI search is most accurate?
None is reliably most accurate on every topic. Accuracy depends more on the question and sources than the tool. Always open the citations for important facts.
How do I turn on Google AI Mode?
Sign in to a personal Google account, search on google.com and tap AI Mode below the search box, or visit google.com/ai. Google’s help page says you need to be 18 or older and not in Incognito, and Workspace accounts are not supported.
Why does ChatGPT sometimes answer without sources?
It decided the question could be answered from its own knowledge, so it did not search. Start your prompt with “Search the web for” or ask it to cite sources, and it will look up current pages.
Sources
- Suprmind: Perplexity pricing and plans (September 2026)
- eesel AI: Perplexity Comet pricing 2026
- TechCabal: Google I/O 2026 AI Mode features (May 20, 2026)
- Big Technology: ChatGPT search is about 20% of Google’s (November 2026)
- PPC Land: ChatGPT sends 190x less traffic than Google (Ahrefs study)
- Wikipedia: ChatGPT Atlas
- Google Search Help: AI Mode (checked November 5, 2026)
