AI can answer a question in seconds. But getting an answer and understanding it are not the same thing. Here is what the research tells us about AI, search, and the thinking we still need to do ourselves.
By Rasheed Akorede Adigun | October 26, 2026
Think about the last time you searched for something online. Did you open a few sources and compare what they said, or did you read the AI-generated answer at the top and move on?
I understand the appeal. We are busy, and sometimes we just need a straightforward answer. I use digital tools to make work easier, too. But studying Library and Information Science and working in an academic engineering library have made me think more carefully about what happens between asking a question and deciding that an answer is reliable.
When I help with research questions, finding information is only part of the work. We also need to think about where it came from, what it means, and whether it actually answers the question.
Now AI can do much of the first part for us. It can summarize a topic, suggest sources, and present a conclusion before we have read a single article.
That raises a question I think students, researchers, librarians, and everyday internet users should be asking: If AI gives us the answer before we search, are we still doing the thinking that helps us learn?
The short answer: AI does not automatically weaken critical thinking. It can help us ask better questions and understand difficult ideas. The risk comes when we stop checking evidence, considering alternatives, or working through problems ourselves because an AI response sounds complete.
What changes when an answer comes before the search?#
In a traditional research process, you might start with a question, search a catalog or database, open several sources, and gradually arrive at a conclusion. Of course, not everyone searches that carefully, and traditional search engines have their own biases.
AI-assisted search can change the order. It often offers a polished explanation first. The sources, if there are any, sit underneath that explanation.
That difference matters. Once we see a confident answer, it is tempting to treat further searching as optional. And the clearer the explanation sounds, the easier it may be to forget that the system could have missed something.
In librarianship, we talk about information literacy: knowing how to find, evaluate, use, and create information responsibly. The Association of College and Research Libraries (ACRL) Framework for Information Literacy reminds us that research is an ongoing process of inquiry and that searching involves exploration, not simply collecting answers.
To me, that is the part we need to protect. The first answer should help us begin thinking, not tell us when to stop.
Are people really searching less when AI summaries appear?#
There is some evidence that answer-first search changes what people do next.
In a 2025 Pew Research Center study, researchers analyzed Google searches from 900 U.S. adults. People clicked a traditional search-result link in 8% of visits when an AI summary appeared, compared with 15% when no AI summary appeared. They clicked a source link within the AI summary in only 1% of visits to pages showing those summaries.
That finding is worth paying attention to, but we should not misread it. Fewer clicks do not prove that people have lost their critical-thinking skills. Someone checking a simple date or definition may genuinely have everything they need.
For a literature review, a health question, or an important personal decision, though, the stakes are different. If the answer is wrong or incomplete, not opening the sources can leave us with a mistaken impression of what the evidence says.
What research says about AI and critical thinking#
I do not think it is helpful to claim that AI is making everyone lazy or less intelligent. The findings so far are more specific, and more useful, than that.
Confidence in AI can affect how much we question its output#
A 2025 CHI conference study by Lee and colleagues surveyed 319 knowledge workers about 936 examples of using generative AI at work.
The researchers found that people who had more confidence in AI reported less critical-thinking engagement, while people with greater confidence in their own ability reported more. They also found that critical-thinking work was shifting toward checking AI output, integrating responses, and overseeing the final result.
This was a study of self-reported experiences, not proof that AI causes a permanent decline in thinking ability. Still, it tells us something important: trusting a tool too easily may change how carefully we use it.
Completing a task is not always the same as learning#
Another 2025 study published in Proceedings of the National Academy of Sciences tested AI tutoring with nearly 1,000 high school mathematics students.
Students using unrestricted AI assistance performed better while practicing. But when that assistance was removed, their performance on an unassisted exam was 17% lower than the control group’s performance. An AI tutor designed to provide hints instead of giving away solutions largely avoided that negative outcome.
The study took place in a particular mathematics setting, so we should not apply the result to every subject or every AI tool. What interests me is the distinction it illustrates: doing well with help is not necessarily evidence that we can do the work independently.
There is a name for letting a tool carry some of the mental work#
Researchers call this cognitive offloading. As Risko and Gilbert explain in their 2016 review, we regularly use external aids to reduce mental demands. A calendar remembers appointments. A calculator handles arithmetic. A notebook stores details we would otherwise need to remember.
There is nothing automatically wrong with that. The question is what we are offloading. Using AI to organize notes may free up time for analysis. Letting it decide which evidence to believe, without checking, is a different matter.
Three habits we risk losing when we accept the first answer#
1. Checking where a claim came from. A reference can be real while the statement attached to it is misleading. A research article might discuss a limited sample, but an AI summary may present its findings as if they apply to everyone. The only way to know is to check the source.
2. Looking for disagreement. Research is rarely a single, neat answer. Scholars use different methods, reach different conclusions, and sometimes disagree about what the evidence means. A smooth AI summary can make those disagreements harder to notice.
3. Explaining something in our own words. Reading a convincing answer can feel like understanding. Yet if we cannot explain the reasoning without the tool, we may only be familiar with the answer, not the subject.
These habits matter beyond university assignments. We use them when judging news, comparing products, reading public health information, or making decisions at work.
How to use AI without giving up your judgment#
I am not arguing that we should return to doing everything manually. AI can be genuinely helpful, especially when a topic is unfamiliar or a research task feels overwhelming.
The difference is in the questions we ask it.
Instead of “Write my argument for me,” try “What assumptions am I making in this argument?”
Instead of “Give me the answer,” try “Give me a hint, then let me work through the problem.”
Instead of “Summarize these studies and tell me what to believe,” try “Where do these studies agree, where do they differ, and what should I verify in the originals?”
These prompts are not a guarantee of accuracy. They simply keep the user involved in the reasoning.
A six-step check before trusting an AI answer#
Here is a practical routine I recommend. I call it SEARCH, as a reminder that receiving an answer is not the end of searching. This is my suggested checklist, not a scientifically validated assessment tool.
- S: State your question. Decide what you actually need to know before asking the tool.
- E: Examine the response. Identify the main claims and anything that sounds too broad or certain.
- A: Audit the evidence. Check important figures, quotations, and citations. Does the source actually support the claim?
- R: Read the originals. For academic work, go beyond the summary and inspect the methods, findings, and limitations.
- C: Compare perspectives. Look for another credible explanation or evidence that challenges the first answer.
- H: Hold on to your judgment. Write your conclusion in your own words, including what you still do not know.
You will not need all six steps every time you ask for a simple fact. But when the question affects your research, learning, or an important decision, they are worth the effort.
An example: Does AI improve student learning?#
Imagine asking an AI tool this question and receiving a confident “yes,” followed by examples of personalized feedback and faster homework completion.
That may be part of the story. But I would want to ask a few more questions: What does improve learning mean? Higher scores while using the tool? Better understanding? The ability to solve a new problem without help?
The mathematics experiment discussed earlier offers a useful caution. Students’ practice performance improved with AI, yet one group’s later independent performance was worse.
A more careful conclusion would be: AI can support learning, but better performance during AI-assisted work does not always mean stronger independent understanding. How the tool is designed and used matters.
That answer is less dramatic. It is also more faithful to the evidence.
What should librarians and educators do?#
I see a real opportunity for academic libraries here. Librarians have always helped people move beyond finding a source to evaluating whether it is useful, credible, and appropriate for a particular question.
That role becomes even more relevant when an answer can be generated instantly.
The ACRL AI Competencies for Academic Library Workers, approved in 2025, emphasize curiosity, skepticism, judgment, responsibility, and the evaluation of AI tools and their outputs.
In practice, I think we should spend more time helping students compare AI summaries with original articles, examine the evidence behind citations, and practice explaining their findings without AI assistance.
Educators can help by designing assignments that value the research process, not just the finished response. And AI developers can help by making sources, uncertainty, and opportunities for verification more visible.
The goal is not to make research unnecessarily difficult. It is to make sure our tools help us learn rather than quietly taking over the part of learning that matters most.
Frequently asked questions#
Does AI make people less intelligent?#
There is no sound basis for saying that using AI automatically makes people less intelligent. Research has identified risks of overreliance and reduced effort in some tasks, but those findings should not be confused with a demonstrated general decline in intelligence.
Can AI improve critical thinking?#
Yes, it can support critical thinking when people use it to test an argument, explore alternative explanations, ask for hints, or identify questions that need further investigation. Users still need to assess whether the tool’s suggestions are accurate.
Should students avoid AI when doing research?#
Not necessarily. AI can help with brainstorming, understanding unfamiliar concepts, and refining search questions. Students should follow course rules and verify important claims against original sources, particularly in academic writing.
Is an AI citation enough to trust an answer?#
No. A citation is a way to point to evidence, not proof that a claim is correct. Open the source and check what it actually says, including any important limits on the findings.
Final thoughts: Keep asking questions#
I value technology that makes information easier to find. I also know, from studying and working in libraries, that access to information is only one part of becoming informed.
We still need to decide what is relevant, what is credible, what may be missing, and what we can reasonably conclude.
So I am less concerned about how quickly AI can answer a question than about what we do after reading that answer. Do we stop there, or do we ask how it knows?
Use AI to help you search, learn, and think. Just do not let the first answer have the final word.
References and further reading#
- Association of College and Research Libraries. (2025). AI Competencies for Academic Library Workers.
- Association of College and Research Libraries. (2016; revised 2026). Framework for Information Literacy for Higher Education.
- Bastani, H., et al. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26), e2422633122.
- Chapekis, A., & Lieb, A. (2025). Google users are less likely to click on links when an AI summary appears in the results. Pew Research Center.
- Lee, H.-P., et al. (2025). The impact of generative AI on critical thinking. Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems.
- Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688.
