Skip to content
Saturday, November 14, 2026Tech, AI and gear explained in minutes.
Work & Productivity

Five Habits That Make Online Research More Reliable

By Owen Marsh   /   October 16, 2026   /   Updated November 8, 2026
five-habits-reliable-online-research

Quick answer: Leave the page. When a claim matters, open a new tab and find out who is behind it, trace it to the original source, and check any image or AI answer against something you can open yourself. That is lateral reading and the SIFT method (Stop, Investigate the source, Find better coverage, Trace). A quick check takes minutes.

Key takeaways

  • Lateral reading means judging a source by what others say about it, not by what it says about itself.
  • SIFT gives four moves: Stop, Investigate the source, Find better coverage, Trace to the original.
  • Searching the exact claim in quotes shows where it started.
  • Reverse image search shows where else a picture has appeared.
  • AI answers are leads, not sources. Verify every fact and citation.

The five habits at a glance

It is 11 p.m. and a friend has sent a “you won’t believe this” article. You could argue in the group chat for ten minutes, or check for two. The times below are rough planning estimates, except the one-minute figure for investigating a source, which comes from Mike Caulfield.

Habit How Time it takes
1. Read laterally Open a new tab and search the site or author, then see what others say About a minute for a quick check
2. Check author, date and purpose Who wrote it, when, and why; filter searches by date 1 to 2 minutes
3. Trace it to the original Search the exact claim in quotes; follow links to the study, document or photo source 3 to 10 minutes
4. Verify AI answers Confirm each fact and citation in a source you can open 2 to 5 minutes per key claim
5. Keep notes with links Save the link, date and one-line verdict About 30 seconds per source

Habits 1 and 2: Read laterally, then size up the source

Leave the page

The Digital Inquiry Group (DIG) describes lateral reading as a strategy used by fact-checkers and skilled internet searchers to evaluate sources. In a post about a Rowan University librarians’ toolkit, DIG describes it as opening new tabs to investigate an unfamiliar source, which helps you look past easily manipulated features such as a site’s domain. Its companion skill, click restraint, means scanning search results before clicking the first one.

DIG’s impact page reports that in a college nutrition course the share of students who left an unfamiliar site to check it rose from 3% to 77%. That is DIG’s own figure, and the page does not say who measured it.

Mike Caulfield’s SIFT, published in 2019, packages the habit into four moves:

  1. Stop. Do not read or share until you know what you are looking at. Oberlin College Libraries adds: notice your emotional reaction and name the claims.
  2. Investigate the source. Find out who is behind it, including expertise and motives. Caulfield says this takes about a minute. Expertise does not guarantee accuracy, and an interested party is not automatically wrong.
  3. Find better coverage. If the claim matters more than the article, look for stronger reporting and compare sources to see where expert consensus sits.
  4. Trace to the original context. Follow the item back to its origin, because clipping, captions and citations can change its meaning.

Author, date and purpose

Oberlin’s guide boils the source question down to two parts: who created this, and why? A company describing its own product is not neutral, though that does not make it wrong. It is your cue to look for a second opinion.

Dates matter because facts age. A price or rule that was right three years ago may be wrong today. Google’s search operators before: and after: narrow results by date. The University of Utah’s AI guide also offers a five-question memory aid, the CRAAP test: Currency, Relevance, Authority, Accuracy and Purpose.

Habit 3: Trace it back to the original

Search the exact claim

Google Search Help says a word or phrase inside quotes finds an exact match. Paste the one distinctive sentence from the post, in quotes, and see who published it first. The same page lists site: before a domain to limit results to that site and a minus sign to remove a word, with no space between operator and term. Pages that repeat each other are one source, not many.

Reverse image search

Google’s help page for searching with an image lists several routes on a computer: click Search by image on google.com and upload a file, drag a file into the search box, paste an image link, or right-click an image in Chrome and choose Search with Google Lens. Results can include websites containing the image or a similar one, and the page suggests adding a text description to your search. Look for the earliest appearance and ask whether the caption changed.

Go upstream

If an article cites a study, statistic or quote, find it and read the relevant part. Oberlin suggests checking what kinds of sources are cited, whether they are represented accurately, and whether the evidence supports the argument. It also notes that sometimes better coverage does not exist. Picture what it would look like and search for it anyway.

Habits 4 and 5: Verify AI answers and keep notes with links

Treat AI answers as leads

The University of Alberta Libraries warns that generative AI tools can produce citations that look real but are hallucinated. Its steps: search for the title in a library site or search engine (not trusting an AI-generated summary in the results), read the source if it exists to confirm the AI described it accurately, and drop the citation if it does not. The same guide says hallucinations will always happen, including in AI features inside search engines and databases, advises against asking a model to verify its own output, and notes that chatbots may tell you what you want to hear.

The University of Utah adds two points. Confident tone is not evidence, and because AI training data has a cutoff date, time-sensitive claims need checking against current sources. A sensible use is brainstorming search terms, then doing the verifying yourself.

Keep notes with links

Good notes take half a minute per source. Jot the claim, the exact link to the page you read (not the homepage), the date, and a verdict of confirmed, unclear or contradicted. Mark which source is the original and which only repeat it, and add a line for what you still do not know. Naming uncertainty beats hiding it.

A worked example, a checklist and common mistakes

Worked example (hypothetical)

A group chat shares a screenshot saying a popular drink “has been proven” to protect against a serious disease, beside a photo of a smiling “doctor.”

  1. Stop. You like the drink, so you want this to be true. Name the claim.
  2. Investigate. In a new tab, search the site’s name. Say it turns out to be run by a company selling a rival product. That does not make the claim false, but it says to find better coverage.
  3. Find coverage. Search the headline’s key phrase in quotes. Do independent outlets or health agencies report the same thing?
  4. Trace. The article mentions a study. Find it, read who was studied and what was found, and compare that with the headline.
  5. Check the photo. Reverse image search the “doctor.” If it appears elsewhere under another name, that is a strong signal.
  6. Note and decide. Save the links and mark the claim “unverified.” Do not forward it.

Checklist

  1. I paused and named the claim.
  2. I opened a new tab and looked up who is behind the source.
  3. I noted the author, date and purpose.
  4. I searched the exact claim in quotes.
  5. I found and read the original study, document or photo source.
  6. I reverse-searched any image.
  7. I verified every AI-supplied fact and citation in a real source.
  8. I saved links, dates and a verdict.

Common mistakes

  • Staying on the page. An “About us” page is the site describing itself.
  • Counting copies as confirmation. Ten pages repeating one post is still one post.
  • Believing a confident tone. AI can sound authoritative while wrong.
  • Letting the model grade itself. Alberta advises against it.
  • Dismissing anyone with a motive. An interested party is not automatically wrong.
  • Skipping notes. Without the link, next month’s you starts over.

Frequently asked questions

How long should checking take?

Caulfield says quick checks suit most browsing and deeper verification suits serious research. Investigating a source takes about a minute, by his estimate.

Can I use an AI chatbot to fact-check something?

Not on its own. Alberta advises against asking a model to verify its own output. Use it for leads, then confirm in sources you can open.

What if I cannot find the original source?

Treat the claim as unverified and say so. If an AI-supplied citation cannot be found at all, Alberta advises not using it.

How do I check a photo someone sent me?

Run a reverse image search, as described above, and look for the earliest appearance. Then compare the caption you were given with the original context.

Before you decide

You do not need all five habits for every cat video. Pick one this week: open a new tab before you trust an unfamiliar site. If a claim is about health, money or safety, read the original too. “I could not verify this” is a perfectly respectable thing to say.

Sources

Last updated: November 2026. This article is general information, not legal, financial or IT-security advice.

Keep reading

Editor’s picks

Explore all topics ↗

Find your next read

7 topics. More possibilities.