“Does it know what happened yesterday?” That’s the second question I get, usually right after the one about paying. The honest answer: not on its own. A language model is trained up to a certain point and knows nothing after that. Anything an assistant tells you about today comes from a search it runs at that moment, and whether it runs one, how, and what it shows you differs by tool and by tier.
This is part 2 of 6. Part 1 covered the gap between free and paid. This part takes one question: which of the five knows what’s happening today, and how do you see where it came from?
Three things to watch
The first is whether the assistant decides to search on its own. ChatGPT does, according to its own help article: ask about something current and it goes looking without being told. Gemini searches through Google and puts the sources under the answer; when exactly it searches, Google doesn’t say. In Claude you switch search on per conversation with the plus button at the bottom left; leave it off and it answers from memory. Perplexity is built the other way round: there, search is the point and the conversation is the extra. Grok searches the web and X, and the other four don’t mention X anywhere in their documentation.
The second is whether you get to see a source. All five show references now, but not equally prominently. Perplexity puts them in the sentence itself, numbered. Claude’s documentation promises a citation with every answer that relied on search. ChatGPT and Gemini put a “Sources” button under the answer, and sometimes a reference in the text. Grok lists “live citations from primary sources” as a feature.
The third is what free allows. Plain search is in the free tier at all five. The deeper work, a research run that reads dozens of pages for minutes and writes a report, is where the paywall stands. At Claude it’s paid only. Perplexity gives you one such run a month for free, plus three uses a day of its heavier search mode. ChatGPT and Gemini give no number, only “limited” and a counter on screen.
The table
Checked on 6 September 2026. Whatever is listed under free is also in paid; the right-hand column only lists what’s added.
| Searches unprompted | Source visible | Free | Paid adds | |
|---|---|---|---|---|
| ChatGPT | Yes, or via the tools menu | “Sources” button, sometimes an inline reference | Search, limited deep research | More deep research, with an on-screen counter |
| Claude | No, switch it on per chat | Citation with every answer that searched | Search and image search; summarising a long article eats your limit | A research mode that stacks multiple searches |
| Gemini | Through Google, no switch | “Sources” button, also into your own Gmail or Drive if you connect them | Search, deep research with a daily limit and no published number, 18 and over | A higher daily limit |
| Grok | Yes, web and X | “Live citations” according to the site | Search across web and X | Higher limit, several agents at once |
| Perplexity | Always | Inline, numbered | Unlimited basic search, three heavy ones a day, one research run a month | Heavy search per week, research per month, pick your own model |
What I see myself
My own news feed runs through Grok, and not as a chat. On X, three automations run on a free account: two send me international wine and drinks news every day, one sends the Dutch news. I use it for my own education, it gives me ideas for content, and I often use it to look things up and to pass news on to other people. The X side is what sets Grok apart here: the other four don’t mention X at all.
What I don’t do is take anything over without checking it. I check everything, and often twice. For wine I sometimes reach for a book. That sounds old-fashioned next to an assistant that cites its sources, but the citation is exactly where things go wrong: a reference proves a page exists, not that the page says what the assistant claims. ChatGPT writes as much in its own help article: open the source and check that it supports the answer.
I haven’t tested ChatGPT, Claude, Gemini or Perplexity separately for this question. What the table says about them comes from their own documentation, not from a week of following the news.
Caveat: who is actually searching
When ChatGPT searches, it rewrites your question into one or more queries and sends those to a search partner, according to its own help article Microsoft or Shopify. If memory is on, what it remembers about you goes into that rewritten query: if you live in Utrecht and eat vegetarian, that’s what it searches for. Claude searches images through Bing. Gemini searches through Google and, if you allow it, through your own mail and files. Every search is also a message sent outward, and you only half control what’s in it. Part 4 is about that.
The second caveat is the source itself. An assistant that searches reads whatever the web hands it, and the web is full of pages written by another assistant. The more you lean on the citation, the more it matters to reach for the book now and then.
What to do tomorrow
Ask your assistant something about today and click the source it puts underneath. Read that page until you find the sentence that carries the answer. If you can’t find it, you now know what a citation is worth.
Frequently asked questions
Can I have an AI follow the news for free?
Yes. Search is in the free tier at all five. If you want an automatic daily round-up, Grok on X can do that, on a free account too.
Why does Claude sometimes say it doesn’t know?
Because search in Claude is on or off per conversation. Switch it on with the plus button at the bottom left of the chat and it will search.
Sources: search in ChatGPT, deep research in ChatGPT, web search in Claude, research in Claude, Deep Research in Gemini, sources in Gemini, Grok, Perplexity plans. Checked 6 September 2026.