Checking what you read online: sources, ads, photos, voices and AI answers
Stop before you share or pay, check who is behind a claim by leaving the page, tell ads from results, trace a photo, and check an AI answer outside the chatbot.
Start here Essentials
At minimum, check one surprising thing you saw this week before you act on it. Everything marked Go deeper can wait.
- Stop before you share it, pay or reply when something surprises you, worries you or asks you to act fast (stop first).
- Open a new tab and search for the site or the person's name, not the claim, to learn who is behind it (leave the page and how to open a tab and come back).
- Look for two other reliable sources on the claim itself (find better coverage).
- For a photo, run a reverse image search to see where else it appeared (reverse image search).
- For your bank, a government site or a download, type the address or use a bookmark instead of clicking a search ad (fake ads at the top).
Minimum time: about 20 minutes. Done when: one claim you saw this week is checked against two outside sources, and what you found is written in your notes.
1. Stop, then check the source somewhere else Understand
Your starting list, Getting back in, Passwords, Two-step sign-in and passkeys, Your phone and tablet, Your computer and home Wi-Fi, Backups and Recognizing scams set up your accounts and devices and taught you to hang up on a caller who wants money or a code. This chapter covers the other way people get moved to act: something they read. It could be a post a friend shared, an ad in a video, a photo in a group chat, a "news" page about a supplement, or an answer from a chatbot. The method is the same for all of them, and most checks take a minute or two.
The method has two sources. The first is SIFT, four moves published by Mike Caulfield, a researcher who teaches web checking: Stop; Investigate the source; Find better coverage; Trace claims, quotes and media back to the original context (SIFT (The Four Moves), Mike Caulfield's blog, 2019). The second is three questions from the Digital Inquiry Group, a nonprofit that grew out of the Stanford History Education Group: "Who's behind the information? What's the evidence? What do other sources say?" (Civic Online Reasoning, Digital Inquiry Group). The four moves are what you do. The three questions are what you are trying to answer. One practitioner's method
| Move | What you do | Question it answers | Usual time |
|---|---|---|---|
| Stop | Pause before you read on, share, pay or reply. Ask whether you know the source. | Do I know what this is? | A few seconds |
| Investigate the source | Leave the page and search for the site, the account or the person's name. | Who's behind the information? | About a minute |
| Find better coverage | Search for the claim itself and look for reliable sources that report it. | What do other sources say? | A few minutes |
| Trace to the original | Find the first place a photo, quote or study appeared. | What's the evidence? | A few minutes |
Stop first
Caulfield's first move has two parts. Before you read or share something, ask whether you know the website or the account and what its reputation is. His wording: "Don't read it or share media until you know what it is." Later, if you notice you have followed one link after another for twenty minutes, stop again and remember what you set out to check.
The stop matters most when the thing in front of you asks you to act: share this now, buy before midnight, call this number, send money to help. Recognizing scams taught the same pause for calls and texts. A post can carry the same pressure as a caller. Nothing about a claim gets truer because it says to hurry.
The stop does not need to be long. Caulfield says depth should match your purpose. If you only want to repost a story or understand an idea, finding out whether the publication is reputable is, in his words, "probably good enough." If you are about to pay, change a medicine or forward a warning to everyone you know, take the extra minutes and go through all four moves. One practitioner's method
Leave the page: lateral reading
The natural way to judge a website is to stay on it: read the About page, look at the logo, scroll through the articles, see whether it looks professional. Researchers call that reading vertically. Professional fact checkers do something else. They glance at the page, then leave it and open new tabs to find out what other people say about the site. This is lateral reading: judging a source by reading about it on other sites, side by side, instead of reading down the source itself.
The term comes from a Stanford study by Sam Wineburg and Sarah McGrew. They watched 10 professional fact checkers, 10 historians with PhDs and 25 Stanford undergraduates evaluate live websites. The historians and students stayed on the sites and were often taken in by features that are easy to fake, such as official-looking logos and domain names. The fact checkers left the sites and reached better-supported conclusions "in a fraction of the time" (Wineburg and McGrew, Lateral Reading and the Nature of Expertise, Stanford Digital Repository).
One task shows the gap. Participants read an article on minimumwage.com and judged who was behind it. Without being prompted, the fact checkers found in an average of 51 seconds that the site belonged to the Employment Policies Institute, and all of them traced it further, to a Washington public relations firm. The historians took an average of 3 minutes 40 seconds, and 6 of the 10 needed a prompt. The people who did best were not the most educated. They had a habit: leave the page and check it elsewhere.
Princeton University Library's guide turns the habit into four steps: open a new browser tab; see what other sources say about the website; find out who runs it; then check the claims and the evidence (Princeton University Library, Lateral Reading guide). The guide also notes that a site's own About page is written by the site. Read it to learn who the site says it is, then check that claim on other sites.
What to type in the new tab: the name of the website, the account or the organization, not the headline. Searching the headline mostly finds copies of the same story. Searching the name finds what other people say about who publishes it. A useful addition is a word such as "funding", "owner" or "review". The Princeton guide gives one more trick for a computer: add -site: followed by the site's address, which leaves the site's own pages out of the results. Product-specific
How to open a new tab and come back
Lateral reading asks you to open a second page and then return to the first. No link in this manual opens a new tab by itself, so you choose when to do it. A tab is one page open in your browser; you can have several open at once and switch between them. If you have not used tabs before, try it once on this page before you need it.
We did not find a maker's help page that we could check today for these steps, and the buttons move between versions. So the table describes what to look for rather than an exact path. Test it
| Device and browser | To open a new tab | To come back |
|---|---|---|
| iPhone or iPad, Safari | Look for Safari's tabs button, the one that shows all the pages you have open. Tap it, then tap the option to add a new page. A new, empty page opens with a search box. | Tap the tabs button again. The pages you have open show as small pictures. Tap the one you were reading. |
| Android phone or tablet, Chrome | Look for Chrome's tabs button, a small square, often with a number in it, near the address bar. Tap it, then tap the option to add a new tab. | Tap the tabs button again and tap the page you started on. |
| Windows, Mac or Chromebook, any browser | Look along the top of the browser window, beside the last tab. Most browsers show a small plus sign there; click it. The browser's menu also has a new tab option. | Click the first tab along the top of the window. Each tab shows the page's name. |
| Any device, without tabs | Write the site's name on paper. Then go to your search page and search for the name. | Use the browser's Back arrow until the first page returns, or open the page again from the post or message. |
The last row is a real option, not a lesser one. What matters is that you read about the source somewhere other than the source. The tabs only save time.
Look before you click: click restraint
When the search results appear, do not open the first one. The fact checkers in the Stanford study scanned the whole list first to see what kinds of sites came up, and only then chose one. The researchers call this click restraint: reading the list of search results before clicking any of them. The first result can be an ad (section 2 covers how to tell), or a site that pays to rank well. A quick scan shows whether a news outlet, an encyclopedia, a government agency or a watchdog group has written about the site.
Find better coverage
Once you know who is behind a post, the next question is whether the claim itself holds. Caulfield's advice: when you care about the claim more than the article that brought it to you, set that article aside and search for the claim, looking for reliable reporting or for what several sources agree on. You do not have to accept what most sources say. You do need to know what it is before you share something that disagrees.
For this manual, the default is two other reliable sources that you found yourself, not ones linked from the post. Reliable here means a source with a name, a known publisher, and a record you can check: a news organization with a corrections policy, a government agency that handles the subject, a university, a large professional or nonprofit group. MedlinePlus, from the National Library of Medicine, gives the same list of good sources for health information and adds a step: once one trustworthy site says something, see whether other reliable sites say the same (MedlinePlus, Evaluating Health Information). Our convention
If you find nothing, that is a result too. A dramatic claim that no news organization or agency has reported is one to hold, not to share.
Trace it to the original
Much of what circulates online has lost its context. Caulfield lists the common forms: a video clipped so it seems to say something else, a real photo with a caption from a different event, a study described as finding more than it did. Princeton's SIFT page asks the same questions for Trace: was the media edited, clipped misleadingly, or real but applied to the wrong situation? The fix is to go back to the first place it appeared and read it there. For a quote, search the exact words in quotation marks. For a study, find the study and read its summary. For a photo, use a reverse image search (section 3).
Checking is not distrusting everything
The goal is to trust the right things, not to doubt everything. Researchers at Stanford evaluated a free one-hour course from Poynter's MediaWise that teaches lateral reading, click restraint and reverse image search. People who took it rated six headlines before and six after, half true and half false. Their accuracy rose from 64 percent to 85 percent, and accuracy on the true headlines rose more than on the false ones, from 52 to 88 percent (Moore and Hancock, Scientific Reports, 2022). The authors call this "discriminant trust": better judgment of true news as well as false. They note that people in the study were already better at spotting false headlines than true ones, and that training should say plainly that not all news is false.
The study's design limits what it can prove, which the Go deeper block below covers. Its takeaway for this chapter is about method: checking a source helps you believe the true thing as much as it helps you doubt the false one.
What the two studies did, and what they cannot show
The fact checker study. Wineburg and McGrew's study had 45 participants doing think-aloud tasks on live sites about bullying, the minimum wage and teacher tenure. In one task, every fact checker judged the American Academy of Pediatrics the more reliable of two pediatrics sites; most students chose the American College of Pediatricians, a much smaller group; and historians often called both reliable. The authors say their study is exploratory and small, ran in an artificial setting with time limits, and might give different results with other sites. It shows what skilled checkers do. It did not test whether teaching that habit works.
The MediaWise evaluation. Moore and Hancock compared 143 people who chose to take the MediaWise course with 238 people from a survey panel who did not. The comparison group's accuracy barely moved, from 55 to 57 percent. The authors list the limits themselves: people were not assigned at random, those who signed up were likely more motivated, the test came right after the MediaWise course so we do not know how long the gain lasts, and the design cannot say which part of the MediaWise course worked. Read it as a short course followed by a large gain in a motivated group, not as proof that one hour fixes anything.
Further reading. MediaWise still offers the one-hour course the study evaluated: How to Spot Misinformation Online, a free one-hour course from Poynter's MediaWise. Caulfield and Wineburg wrote a book on the method, Verified: How to Think Straight, Get Duped Less, and Make Better Decisions about What to Believe Online (University of Chicago Press, 2023); a library review is in LOEX Quarterly, volume 52, issue 2. We list the book as further reading; we have not summarized it here.
2. Ads, search results and online shops
Some of what looks like news or a search result is paid for. That is legal when it is labeled. The trouble is that labels are small, and some sellers and scammers count on readers missing them. This section covers what an honest label says, where fake ads show up, and how to reach your bank, a government site or a download without going through an ad.
Ads that look like news
Native advertising is an ad made to look like the articles or results around it. The Federal Trade Commission (FTC) has a guide for businesses on how such ads should be labeled. It lists "Ad", "Advertisement", "Paid Advertisement" and "Sponsored Advertising Content" as labels people are likely to understand. It calls "Promoted" and "Promoted Stories" "at best ambiguous", and says a label belongs in front of or above the headline (FTC, Native Advertising: A Guide for Businesses, December 2015). The guide is written for businesses, so it tells you what an honest label looks like. It does not mean every site follows it. Convention
| Label you see | What it usually means | What to do |
|---|---|---|
| Ad, Advertisement, Paid Advertisement, Sponsored | Someone paid to put it there | Treat it as a seller's message. Check the seller as you would any source. |
| Promoted, Promoted Stories | Often paid; the FTC guide calls the word ambiguous | Assume it may be an ad until you check. |
| Presented by, Brought to you by, Sponsored by | A sponsor paid; it may or may not have written it | Look up the sponsor in a new tab. |
| A box called "From around the web" or "More content for you" | The FTC's example of ad links that readers take for news | Scroll past it, or check each link's source first. |
| A company logo with no words | The FTC guide says a logo alone does not tell readers it is an ad | Assume it may be an ad until you check. |
Missing a label is not a reader's failing. FTC staff tested 48 people in a lab on search pages and news sites, on a computer and an Android phone, with eye tracking. Clearer labels raised the chance a person recognized an ad by 10 to 45 percentage points. Even so, the report says a significant share of people still did not recognize some ads as ads (FTC staff report, Blurred Lines, December 2017). The research is from 2014 and 2015, and page designs have changed since. Its point stands: ads near the top of a page are easy to take for results, for careful readers too.
Fake ads at the top of search results
The FBI warned in December 2022 that criminals buy search ads that pretend to be real businesses. The ads appear at the very top of the results, the FBI says, "with minimum distinction between an advertisement and an actual search result." They lead to copies of real websites: fake bank and cryptocurrency sign-in pages, and downloads that install harmful software (FBI Internet Crime Complaint Center, public service announcement on search engine ads, 2022).
The FBI's tips: before you click an ad, check the web address for typos or a misplaced letter, and rather than searching for a business or a bank, type its address into the browser's address bar. This manual's default follows from that. For your bank or card, a government site such as the IRS or Social Security, and any software you download, type the address yourself or use a bookmark you saved from a page you typed. Get the address from your card, your statement or a letter. Do not get it from an ad. Our convention
A bookmark is a saved link to a page, kept in your browser so you can open it again without searching. Save one for your bank the next time you type its address. The FBI also suggests an ad-blocking browser extension for searches. This manual does not recommend extensions or products; typing the address works on every device without one.
A phone number in a search ad or a search result can belong to a scammer, the same as a link. For your bank, card or phone company, use the number on your card or statement. Recognizing scams covers the call-back habit, and the who to call table lists the official reporting lines.
Google's built-in shortcuts
Google Search has two tools that do a quick lateral read for you. Neither replaces your own search, but both are fast.
- About this result. On a Google search result, select More (three dots) after the web address. The panel can show what Google knows about the source, such as when Google first saw it or a Wikipedia description. Google notes that the feature is only available in some regions (Google Search Help, Learn more about a web page). Product-specific
- About this advertiser. On a Google ad, select More or Info, then About this advertiser. Google says it requires advertisers to verify their identity with legal documents showing their name and location (Google My Ad Center Help, About advertiser verification). Verification tells you who placed the ad. It does not tell you the offer is honest, and the FBI's warning above shows fake brand ads still run. Product-specific
Online shops you have not used before
A shop that appears in an ad or a post can be a copy of a real brand's site, or a site that takes the money and ships nothing. The FTC's consumer page on online shopping suggests searching the company or product name with the words "complaint" or "scam" and reading what other people say about their experience (FTC, Online Shopping, last updated May 2023). That is lateral reading for a shop.
The same page says that "https" at the start of an address means the connection is encrypted, not that the site is legitimate, since scammers can encrypt sites too. Browsers often show a padlock for the same thing; MedlinePlus describes the lock and https as signs of a secure connection. A secure connection means other people on the network cannot read what you send. It says nothing about who is on the other end. Physical principle
3. Photos, video and voices
A photo can be real and still mislead: a flood from another year, a crowd from another city, a person who is not who the caption says. A photo or a voice can also be made by software. This section starts with the check that works for both, finding where an image came from, and then covers what is different about AI images, video and voices.
Reverse image search
A reverse image search uses a picture, instead of words, as the search. The results show other places the same or a similar image appears online. For checking, the part that matters is the list of pages that use the image: if it appeared years before the event the post describes, or on a page about something else, the caption is wrong.
Google's tool for this is Google Lens. Its help page says the results can include an AI summary, results for objects in the picture, similar images, and websites with the image or a similar image (Google Search Help, Search with an image on Google; the same page has iPhone and iPad and Android versions). Scroll past the summary to the websites. Product-specific
First, save the photo or take a screenshot of it, so it is in your photos. Then follow the steps for your device. Your green box in Your setup shows only your own devices.
iPhone or iPad
- Open the Google app, the Chrome app or Safari.
- In the search bar, tap Google Lens.
- Tap Photo picker.
- Choose the photo or screenshot.
- Scroll down to Websites with the image.
In short: Google app > Google Lens > Photo picker. For a picture on a web page in the Google app or Chrome, touch and hold the picture, then tap Search inside image (Google app) or Search image with Google (Chrome). Google says Chrome needs Google set as its search engine for this.
Android phone or tablet
- Open the Google app or Chrome.
- In the search bar, tap Google Lens.
- Under Screenshots, choose the photo.
- Scroll down to Websites with the image.
In short: Google app > Google Lens. For a picture on a web page in the Google app or Chrome, touch and hold it and tap Search image with Google Lens. Pixel shown; other makers' menus differ.
Windows, Mac or Chromebook
- In your browser, go to google.com.
- Click Search by image.
- Click Upload a file.
- Choose the photo and click Open or Choose.
- Look for Websites with the image.
In short: google.com > Search by image > Upload a file. Google lists Chrome, Firefox, Safari and Edge as browsers that work. In Chrome you can also right-click a picture on a page and click Search with Google Lens.
Microsoft's Bing does the same on any device: on Bing Images, select the camera icon in the search box, then upload a photo or paste its address (Microsoft Support, Using Bing Visual Search). Google notes that it may store the addresses you search, and Microsoft says photos you provide may be used to improve its image service. Use a screenshot that shows only the picture, not your messages. Product-specific
When did the photo first appear?
The question a reverse image search answers best is "is this photo older than the event it claims to show?" TinEye, a reverse image search site, is built for that. It says its index is smaller than Google's, but that it does not delete old results and shows when each match was found (TinEye Help, Differences between TinEye and Google Search By Image). You upload a photo or paste its address on the TinEye website; it has no phone app, and the site works in a phone's browser. TinEye says it does not add uploaded photos to its index and keeps them only briefly (TinEye Help, How do I search on TinEye?). Product-specific
No result is not a verdict. TinEye says it does not typically add social media images to its index, because most social media sites prohibit it (TinEye Help, About the TinEye index), so a photo that only ever appeared on social media may show no matches. A missing match means you have not found the original yet. It does not mean the photo is new, and it does not mean it is fake.
AI images, video and voices: verify first
Generative AI is software that makes new text, pictures, audio or video from examples and instructions. The FBI's December 2024 announcement lists how criminals use it: realistic profile photos, fake IDs, pictures of celebrities promoting counterfeit goods, pictures of disasters for fake charities, short clips of a relative's voice in a fake emergency, and videos of public figures for fake investment promotions (FBI Internet Crime Complaint Center, Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraud, 2024).
The FBI's 2025 annual report counted 22,364 complaints that mentioned AI. It says AI-made content is "becoming increasingly difficult to detect and easier to make" (FBI, 2025 IC3 Annual Report (PDF), AI section). Those are complaints people chose to report with an AI detail. They count reports, not every case.
So the order of checks matters. Verify where the thing came from first; look for flaws second.
- A voice on the phone asking for money: hang up and call the person back at a number you already have, or ask for your family code word. Recognizing scams covers both. The call-back works whether the voice was cloned or not.
- A video of a famous person recommending an investment or a product: search the product's name with the word "scam" in a new tab, and go to the person's own website or official account to see whether they said it there. See Frank's example in See it.
- A photo: run a reverse image search and look for the first place it appeared.
- Then, and only as a second step, look closely. The FBI lists signs such as distorted hands or feet, odd teeth or eyes, unrealistic shadows, lag, a voice that does not match the mouth, and unnatural movement. Princeton's library guide adds blended details and too-smooth textures, and says these signs do not guarantee anything (Princeton University Library, Generative AI guide).
The FBI also suggests limiting the public photos and voice recordings of yourself online and making social accounts private. That reduces what a scammer can copy. It does not help with a clip that is already public, which is why the call-back comes first. Convention
No sign proves a picture, video or voice is real or fake. An odd-looking hand can be a bad photo. A clean image can be made by software. A missing credential, an empty reverse image search or a detector's score proves nothing either way. What you can find out is where the thing came from and what reliable sources say about it. If you still cannot tell, do not act on it and do not share it.
Why we do not suggest AI detectors
Websites that promise to tell whether a text or picture was made by AI are easy to find. Princeton's library guide says AI detection software is "highly inaccurate." It notes that detectors have labeled historical documents as AI-written, miss text that was made by AI, and flag writing by people who learned English as a second language. A detector's score is one more claim to check, not an answer.
Content Credentials: what they record and what they cannot
Content Credentials are a record attached to a photo, video, audio file or document that says which camera or app made it and what edits were made. They come from the Coalition for Content Provenance and Authenticity (C2PA), a standards group whose members include camera makers, software companies and news organizations. The record is signed, so a change to the file or the record breaks the signature and shows that something was altered. Some viewers show a small icon when a file has a credential (C2PA, Frequently asked questions).
The standard itself is clear about the limit. Its explainer says provenance information "alone cannot tell you whether the digital content is true, accurate or factual" (C2PA, Content Credentials Explainer, version 2.4). A credential can tell you a photo came from a certain camera on a certain day. It cannot tell you the caption describes what the photo shows. C2PA's FAQ also says credentials can be removed from a file, and its answer is a fallback system it calls Durable Content Credentials. Most pictures you see online carry no credential, so a missing one is not a sign of anything.
4. Chatbot answers, health claims and fact checks
Three kinds of source need their own notes: answers written by AI, claims about health, and fact checks. The method is still lateral reading. What changes is where you look.
Chatbot answers and AI search summaries
A chatbot is a program that answers questions in conversational sentences, such as the AI assistants built into search pages, phones and apps. The usual first check, who wrote it, does not work on a chatbot's answer. UC Irvine's library points out that the answer has no author and the system works like a "black box" (UC Irvine Libraries, Generative AI and Information Literacy). Answers can sound sure of themselves and be partly right. A made-up fact in an AI answer is called a hallucination.
Centre College's library gives steps that fit the rest of this chapter (Centre College, Grace Doherty Library, Evaluate information found using AI):
- Break the answer into claims. Pick out each specific fact you could search for: a number, a date, a name, a rule.
- Check each claim outside the chatbot. Open a new tab and search for it, as in section 1.
- Open every source the answer cites. Confirm the page or article exists and says what the answer claims. The library warns that chatbots can make "very convincing fake citations", sometimes with real researchers' names.
- Go to the organization that should know. For a benefit, the agency that pays it. For a tax rule, the IRS. For a product recall, the maker or the agency that issued it.
The same applies to AI summaries at the top of search results. Google calls its version an AI Overview and says on its help page: "AI responses may include mistakes" (Google Search Help, AI Overviews). Treat the summary as a lead. Open the linked sources and check the claim there. Product-specific
Health claims
Health claims are where acting too quickly can hurt the most. MedlinePlus gives two key steps: ask questions before you trust what you read, and discuss what you find with your health care provider before you rely on it. Its questions are a lateral read for health:
- Who runs the site? Look for the owner on its About page. A site with no owner listed is a warning sign.
- What is its purpose? To inform, to sell or to promote an opinion?
- Who pays for it? Ads should be clearly marked. MedlinePlus says: "Watch out for ads designed to look like neutral health information."
- Is it high quality? Be wary of dramatic writing, promises of a cure, and claims that sound too good to be true.
- Where does the information come from, and is it up to date? Look for listed sources and for a date it was written, reviewed or updated.
The National Center for Complementary and Integrative Health (NCCIH), part of the National Institutes of Health, describes a common trap: fake news sites that sell health products. They may copy real news outlets' logos, use similar names and addresses, and describe an "investigation" of a product. Its warning signs include a "news" page that endorses a product, a shocking headline, research that sounds too good to be true, quotes only from satisfied people, links to a sales site, and only positive comments with no way to add your own (NCCIH, Are You Reading News or Advertising?).
The Food and Drug Administration (FDA) defines health fraud as products that claim to prevent, treat or cure diseases or other health conditions "but are not proven safe and effective for those uses." The FDA says they waste money, can delay proper diagnosis and treatment, and can cause serious or even fatal injuries (FDA, Health Fraud Scams).
This manual gives no medical advice and does not judge whether any health claim is right. Checking a source tells you who is behind a claim and what other reliable sources say. It does not tell you whether a treatment, a supplement or a change fits you. Take what you find to your doctor, pharmacist or other health care provider, and do not stop, start or change a medicine or treatment because of something you read online. Your provider decides with you.
Fact-checking organizations
A fact check is an article that looks into one specific claim, says what the evidence shows, and lists its sources. Fact-checking sites are a good place to search when a claim is spreading, because someone may already have traced it. Princeton's guide lists AP Fact Check, FactCheck.org, PolitiFact and Snopes. The International Fact-Checking Network (IFCN), run by the Poynter Institute, keeps a list of organizations that sign its code of principles; when we checked the list on October 19, 2026, all four were verified signatories.
The most useful part of a fact check is not the rating. It is the sources. The IFCN code asks signatories to give their sources in enough detail that readers can repeat the work themselves (IFCN Code of Principles, the commitments). Open the sources in a fact check and read them. A fact check is a starting point, not the final word. Convention
What IFCN signatories commit to, and who the main US fact checkers are
Signatories are assessed against the code by independent assessors, and an organization whose editorial work is controlled by a government, party or politician is not eligible to sign. The first principle is nonpartisanship and fairness: the same standard for every fact check, no concentration on one side, no policy advocacy, and a published statement on how claims are chosen. The second is the sources principle above. Signatory status expires and is renewed, so check the list again when it matters.
- Snopes, About Us: began in 1994 looking into urban legends and hoaxes, now covers news and viral claims; carries ads.
- PolitiFact, Principles of the Truth-O-Meter: owned by the nonprofit Poynter Institute; rates claims by elected officials and others on a six-step scale.
- FactCheck.org, Our Mission: a project of the Annenberg Public Policy Center at the University of Pennsylvania that checks claims by major US political figures; its SciCheck covers science claims.
We describe these organizations; we do not rate them. Whether a given fact check is right is settled the same way as anything else: by reading its sources.
For health information, the National Library of Medicine's Evaluating Internet Health Information tutorial walks through two made-up example sites.
When you are not sure, do not share
MedlinePlus says a post from someone you know is not necessarily good information, and that if you are unsure, you should not share it. That works for every kind of claim in this chapter. Not sharing costs nothing. You can come back to it once you have checked. If you decide it was false, a short, kind reply to the person who sent it, with the source you found, helps more than silence. Scams and false posts can reach anyone; the person who made them is responsible, not the friend who passed them on. Convention
5. See it See it
Denise reads a supplement claim. Denise Okafor, a nurse in Columbus, Ohio, helps her mother Grace set up her iPad on Saturdays. On her Chromebook she sees a post in a group she follows. It links to a page styled like a news site, headlined as an investigation, saying a supplement can replace one of the prescriptions Grace takes. The post, the page and the supplement are fictional.
Denise stops before forwarding it to Grace. She opens a new tab and searches for the site's name, not the headline. The results show no news organization by that name, and a consumer complaint page about the same sales address. Back on the page, she runs through NCCIH's signs: it endorses one product, quotes only satisfied buyers, links to a sales page, and its comments are all positive with no box to add one. On MedlinePlus she looks up the supplement and finds no reliable source that says it replaces a prescription medicine.
Known: the page sells the product it praises; its owner is not listed; Denise found no reliable source that supports the claim. Assumed: that her search found the main coverage that exists. That is her assumption from one search session. Open: whether the claim holds for Grace. This manual does not settle that, and neither does Denise's search. Grace's doctor or pharmacist decides, with Grace.
Denise does not forward the post. She writes the site's name, what she found and the date in her notes, and puts one line on the list for Grace's next appointment: "Ask about this supplement and the prescription." Under these assumptions, the post is a sales page, not news. Whether anything about Grace's medicines should change is a question only her doctor or pharmacist answers.
Frank sees a video ad. Frank Tran, in Sacramento, California, is watching videos on his iPhone when an ad plays. A famous investor appears to recommend a cryptocurrency platform called CoinHarbor Pro and says new users double their money in a month. The ad, the platform and the claim are fictional; the investor is not named here.
Frank stops before he taps the ad. On his MacBook he opens a new tab and searches "CoinHarbor Pro scam." The results include warnings from people who say they could not withdraw money. Then he goes to the investor's own website, typing the address rather than using the ad, and to the investor's official video channel. He finds no mention of the platform. He remembers that the FBI lists AI video made for "fictitious or misleading promotional materials for investment fraud schemes," and that this manual says to verify first and look for flaws second. The lip movement looks slightly off, but that is the second check, not the first.
Known: the investor's own website and channel do not mention the platform; others online report problems with it. Assumed: that the investor's official site is the one Frank found. He typed an address he knew from the news, so the assumption is his. Open: whether the video was made by AI or edited from a real clip. Frank has no way to tell that from the ad, and the answer does not change the decision. Whether the platform is real is settled, in this fictional case, by the searches: it is not.
Frank does not tap the ad and does not send money. He reports the ad through the video app's own report option and writes what he found in his notes. Frank already keeps an incident log from the earlier bank imposter call (When something goes wrong), so he adds a line there too.
Free pages that show the method in the publishers' own words:
- SIFT (The Four Moves): Mike Caulfield's original post, with a short example for each move.
- Princeton University Library, Lateral Reading: the four steps and the
-site:trick. - MedlinePlus, Evaluating Health Information: the questions for a health site, post or news story.
6. Practice Practice
Minimum: one claim you saw this week, checked against two outside sources that you found yourself, with the result written in your notes. Time: about 20 minutes. Materials: your phone, tablet or computer, one post, ad, photo, message or chatbot answer that surprised you, and My notes & bookmarks or a sheet of paper.
- Pick one claim. Choose something you saw this week that surprised you or asked you to act. Write the claim in one sentence.
- Stop. Do not share it, reply or pay while you check.
- Open a new tab. Use the steps in the tab table, or write the name on paper and search for it.
- Search for who is behind it. Type the name of the site, the account or the company. Scan the results before you click any.
- Search for the claim. Find two reliable sources that report on it, or note that you found none.
- If there is a photo, run a reverse image search. Use the steps in reverse image search and note the earliest page you find.
- Write it down. In your notes, write the claim, who is behind it, the two sources (or "none found") and what you decided: shared, not shared, or asked someone who knows.
Run the same photo through TinEye as well as Google Lens and compare the earliest dates each shows. Then take one answer from a chatbot or an AI Overview, break it into three claims, and check each one with the steps in chatbot answers. Note whether every source it cited exists and says what the answer claimed. The check card has the four moves on one printable page.
7. Your setup Your setup
The box below reads the devices you saved and shows the reverse image search and new tab steps for each one.
8. Check your work Check your work
Criteria
- Did you stop before sharing, replying or paying?
- Did you search for who is behind the claim on a page other than the one that showed it to you?
- Do your notes name two outside sources you found yourself, or say that you found none?
- If there was a photo, do your notes say where else it appeared and the earliest date you found?
- Does your note say what is still open, and who could settle it?
Worked example
Ruth checks one photo on her iPhone. Ruth Alvarez, a retired school librarian in Tucson, Arizona, sees a post in her neighborhood group: a photo of a bobcat on the edge of a backyard pool, captioned "Tucson this morning. Keep your pets in." Three neighbors have already shared it. The post, the photo and every result below are fictional.
Ruth stops before sharing. She takes a screenshot, opens the Google app on her iPhone, taps Google Lens, taps Photo picker, and chooses the screenshot. Under Websites with the image she finds the same photo on a news story from another state, dated several years ago, and on two pages that repeat it. She then opens a new tab and searches for bobcat sightings in Tucson. She finds a local TV station's story from last month and the state wildlife agency's page on living near bobcats, both of which she reaches by searching their names herself.
Known: the photo appeared online years before the post; it is not from this morning. Assumed: that the oldest page Lens showed is close to the original. Ruth's assumption, since Lens may not show every copy. Open: where the photo was really taken, which only the original photographer or the outlet that first published it can settle; and whether bobcats visit her own street, which the state wildlife agency can answer.
She writes in her notes: the caption, "photo is older and from another state," the two outside sources, and "did not share." That meets the criteria: she stopped, checked the source somewhere else, found two outside sources, traced the photo and named who settles what is open. Under these assumptions the caption is wrong; the advice to keep pets in may still be reasonable, and the wildlife agency's page is where she reads it.
Common mistakes
- Searching the headline instead of the source. You find copies of the same story. Search the site's or the person's name first.
- Clicking the first result. It may be an ad. Scan the list first.
- Counting links from the post as outside sources. The post chose them. Find your own.
- Treating "no reverse image results" as proof. TinEye does not usually index social media photos. No result means not found yet.
- Trusting a chatbot's citations without opening them. Confirm each one exists and says what the answer claims.
- Acting on a health claim before asking your provider. Write the question down and bring it to your next appointment, or call the pharmacy.
What is still unresolved
This chapter does not settle whether a given claim is true. It shows you how to find out who is behind it, what reliable sources say and where a photo came from. Some claims stay open after a good check: a new event that no one has reported yet, or a study that experts still argue about. In those cases the honest note is "not settled," and this manual's default is not to share. Health questions are settled by your health care provider. A scam you have already paid is covered in When something goes wrong.