AI with Web Search, Deep Research and AI Mode: When the Answer Is Fresh

By Ravindra Bagale, trainer · LinkedIn · Last updated: 19 October 2026

Generative AI course · Chapter 14 · Estimated reading time: 15 minutes

1. Suppose we asked about today's train news

Suppose we are Vikram, travelling from Pune to Delhi tomorrow. We heard that some trains are running late because of track work, and we ask an AI with search off:

Are there any train delays from Pune to Delhi today?

A typical answer (yours will be similar, not the same):

I don't have access to real-time information. Please check the official railway enquiry
website or app for the latest train status.

That is an honest answer: the model alone does not know today. Now we turn web search on (or use a tool that searches automatically) and ask again. This time the app searches the web, reads a few pages, and gives the model fresh text to answer from, with links.

Expected output with search (yours will be similar, not the same):

According to [news site] (updated today), some trains on the Delhi route are running late
because of track maintenance work. [1]
For your exact train, check the live status on the official railway enquiry site with your
train number. [2]

That is the idea of this chapter:

Web search lets the AI read fresh pages before it answers, so the answer can be about today, with links you can check.

2. How AI web search works (in 4 steps)

1. You ask a question.
2. The app searches the web, often with several searches at once for different parts of your question.
3. It reads the top pages and adds the useful parts to the model's context window.
4. The model writes an answer from those pages and shows the sources as links or numbers.

Google describes this for AI Mode as a "query fan-out": it divides your question into subtopics and searches for each one at the same time.

Two things follow from this:

• The answer is only as good as the pages it found. A weak blog in, a weak answer out.
• The model can still misread a page. You learned this in the hallucination chapter: always open the important links.

3. The 3 levels

Where to find them on the free plans (checked October 2026):

4. Worked example 1: today's news, with sources

Steps:

1. Open google.com/ai (AI Mode), or Copilot, or ChatGPT with search on.
2. Paste:

What are the 3 most important business news stories in India today?
For each: one line summary, the source name, and the time or date it was published.
Only use news from the last 24 hours.

3. Open the link for each story.

Expected output (yours will be similar, not the same):

1. [One-line summary of story 1] – Source: [news site], [time, today]
2. [One-line summary of story 2] – Source: [news site], [time, today]
3. [One-line summary of story 3] – Source: [news site], [time, today]

Check it (1 minute):

1. Is each date really today? Search tools sometimes pick an older article.
2. Is the source a known news site?
3. Does the article say what the summary says?

A useful follow-up:

For story 2, explain in 5 simple lines why it matters for a salaried employee.

Now the AI explains (its strength) using fresh facts (from search). This mix is where AI search shines.

5. Worked example 2: a laptop price comparison

Prices change every week, and a model alone often gives old or invented prices. With search, the AI can find current listings, but you still check the final price on the shop's own page.

Suppose we are Pooja, a first-year BCA student. Our budget is about ₹50,000 for a laptop for coding and online classes.

Step 1: ask for the shortlist, not the final answer.

I am a BCA student. Budget: about ₹50,000. Use: coding (Python, Java, VS Code), online classes,
some Excel. Search the web for current laptops in India in this budget.
Make a table: Model | Processor | RAM | Storage | Weight | Price seen | Where seen (with link) | Date of the page.
Show 4 laptops. If you cannot find a current price, write "Price not found".
After the table, say which 2 you would shortlist for my use and why, in 3 lines.

Expected output (yours will be similar, not the same):

| Model     | Processor | RAM   | Storage   | Weight | Price seen  | Where seen        | Date of page |
|-----------|-----------|-------|-----------|--------|-------------|-------------------|--------------|
| [Model A] | [CPU]     | 16 GB | 512 GB SSD| 1.6 kg | ₹[price]    | [shop site, link] | [date]       |
| [Model B] | [CPU]     | 8 GB  | 512 GB SSD| 1.4 kg | ₹[price]    | [shop site, link] | [date]       |
| [Model C] | [CPU]     | 16 GB | 512 GB SSD| 1.7 kg | Price not found | –             | –            |
| [Model D] | [CPU]     | 16 GB | 1 TB SSD  | 1.8 kg | ₹[price]    | [shop site, link] | [date]       |

Shortlist: Model A and Model D. Both have 16 GB RAM, which helps when VS Code, a browser and
a video class run together...

We do not print real model names and prices here because they change every week. Your table will show real ones.

Step 2: check the 2 shortlisted laptops yourself.

1. Open each link. Is it the same model number? (Small letters at the end of a model name can mean a different RAM or processor.)
2. Check today's price on 2 shops' own websites, and the brand's official site.
3. Check the warranty and the exact RAM (some laptops cannot be upgraded later).

Step 3: let the AI explain, not decide.

Compare [Model A] and [Model D] for a BCA student in 5 simple lines: speed, battery, weight, upgrade options, after-sales service.
Do not give prices.

6. Worked example 3: a deep research report on a career

Deep research is for bigger questions, where you would normally open 20 tabs and spend an evening.

Suppose we are Sameer, a B.Com graduate in Indore, thinking about becoming a data analyst.

Steps (Gemini Deep Research; ChatGPT deep research is similar):

1. Open gemini.google.com and choose Deep Research in the prompt bar.
2. Paste:

I am a B.Com graduate in India with basic Excel. I am thinking of a career as a data analyst.
Research and write a report with these sections:
1. What a data analyst does in a normal day (with 3 real job-post examples from India and links)
2. Skills asked for in entry-level job posts in India (Excel, SQL, Power BI, Python, etc.), as a table
3. Free and low-cost ways to learn each skill
4. A realistic 6-month learning plan for 2 hours a day
5. Entry-level salary ranges in India, only from sources that show their method and date
6. Risks and honest challenges for a commerce graduate
Use Indian sources where possible. Give the source for every number.

3. Gemini shows a research plan first. Read it. You can edit it (for example, "add a section on interview questions") before you start.
4. Select Start research and wait. It can take several minutes. You can do other work meanwhile.
5. When it is done, read the report and open the sources for every number.

Expected output (yours will be similar, not the same):

Data Analyst Career Report for a B.Com Graduate in India

1. What a data analyst does
   - Cleans sales or finance data in Excel or SQL every morning... [1]
   - Example job posts: [company/role], [city] – [link] ...

2. Skills in entry-level job posts
   | Skill    | How often asked | Why it matters |
   |----------|-----------------|----------------|
   | Excel    | Most posts      | ...            |
   | SQL      | Most posts      | ...            |
   | Power BI | Many posts      | ...            |
   ...
5. Salary ranges
   - [Source name, date]: ₹[range] per year for entry level, based on [method]...

How to use the report:

1. Check every number in the salary section. Salary figures differ a lot between websites; prefer sources that say how many salaries they used and when.
2. Open 3 job-post links. Are they real, current, and in India?
3. Turn the plan into action: paste section 4 into a new chat and ask, "Make this a weekly checklist I can tick."

Deep research saves hours of searching. It does not remove the need to check. A long report with 40 sources can still contain a wrong number, and a long report is harder to check, so focus on the facts you will act on.

7. Google AI Mode in 2 minutes

AI Mode is Google's AI search experience inside Google Search. It is free and available in India.

How to open it:

1. Go to google.com/ai, or
2. Type a question on google.com and select AI Mode, or
3. In the Google app, tap AI Mode on the home screen.

What you can do:

• Ask long questions in plain language, by typing or by voice.
• Ask follow-up questions in the same thread.
• Add an image or a PDF to your question.
• Paste a web page link and ask about that page.

Good questions for AI Mode:

Compare the metro, bus and auto for a daily 12 km office commute in Pune: time, monthly cost and comfort. Show links.

What documents do I need to renew a passport in India? Give the official link for each step.

Google's own help page says AI Mode does not always get it right, and suggests checking important information in more than one place and opening the links. That advice is exactly the fact-check from the last chapter.

AI Mode vs AI Overviews: an AI Overview is the short AI summary that sometimes appears at the top of normal Google results. AI Mode is a full conversation you can continue, with more searches behind each answer.

8. When search helps and when it does not

Turning search on for a writing task can make the answer worse: the AI copies phrases from web pages and adds links nobody needs.

9. Check a researched answer (the 4 questions)

1. Date: is every source recent enough for this question?
2. Source: is it the official site, a known news site, or a random blog?
3. Match: does the page really say this? (Ctrl + F the key word.)
4. Missing: what did the search not cover? Ask: What important points might this answer be missing?

10. Compare the 4 tools and AI Mode

Limits and features change often. Check each tool's help page for the current limits on your plan.

11. Beginner mistakes and quick fixes

12. Try at home

Recap

You have now finished the "how AI works" module: models, tokens, hallucinations and search. Next time an AI answer looks strange, you will know which of the 4 to blame, and how to fix it.

Previous chapter: Why AI Hallucinates and How to Fact-Check Any AI Answer

Next chapter: Multimodal AI: Ask Questions About Photos, PDFs and Voice Notes

Ravindra Bagale, trainer: linkedin.com/in/ravindra-bagale. The people, trips and laptop rows on this page are made up for learning; no real prices are shown because they change every week. Feature availability for ChatGPT, Gemini, Claude, Copilot and Google AI Mode was checked on the official help and pricing pages on 19 October 2026 and can change. AI answers change every time, so your output will be similar, not the same.