Briefing: New web search tools to cut through the slop

Our weekly Briefing is free, but you should upgrade to access all of our reporting, resources, and a monthly workshop.

Craig shared 25 tips and tools he learned at the Summer School on Investigative Reporting in Riga, Latvia. Highlights include a bot that profiles Telegram users, a catalogue of more than 180 public databases, and a former private investigator's advice on getting strangers to talk.

In the latest episode of Show & Tell, our podcast taking you behind the scenes of digital investigations, we spoke to Wall Street Journal reporters Caitlin Ostroff and Neil Mehta about a deceptive marketing campaign run by Polymarket. The WSJ found that the prediction market made a fake version of its own website and paid creators to film themselves winning fake bets. The episode also features a rare instance of someone saying "I miss Google Plus." Check it out on YouTube or wherever you get your podcasts.

📍 Prediction market Kalshi stopped offering “mention markets” in the sports vertical following a probe by the Commodity Futures Trading Commission. These markets allowed users to bet on whether someone would say a specific word or phrase at a given moment. Such markets are eminently riggable and were featured in the deceptive Polymarket ads exposed by the WSJ article that’s the subject of this month’s Show & Tell.

📍 An AI tool called Kromix ran ads on Meta that promoted its ability to generate realistic nudes of a female US politician. Following Wired’s reporting, the ads were taken down and Apple removed Kromix from the App Store.

📍 South Korean police arrested 419 suspects for “deepfake crimes” in the first seven months of this year, according to SBS.

📍 WhatsApp announced an optional “Scam Alert” feature that scans incoming messages on-device for known scam patterns and alerts the recipient. The feature is in beta for now.

📍 Lead Stories, PolitiFact, FactCheck.org, Factchequeado, and AFP have teamed up to publish a weekly fact-checking newsletter ahead of the US midterms. You can sign up here. (Disclosure: Lead Stories is an Indicator advertiser; this is not a sponsored mention.)

Tara Calishain is a writer and web search expert who builds useful tools like Search Tweaks, MegaGladys, and Local Search America. I did a quick email Q&A with her about some of her latest tools and her perspective on what AI is doing to search.

An edited excerpt is below. You can learn more about her at her personal site and her long-running blog, ResearchBuzz. — Craig

Your personal site has the motto, "I believe you deserve easy paths to useful information in the same way you deserve clean food and water." I love that. How do the tools you build at searchtweaks.com and elsewhere fit into that mission statement?

I have been writing about search engines and web search for over 30 years. In those 30 years web search has gone from lots of tools to explore and search the Web in lots of ways to mostly Google. There are other search engines available of course but Google has the overwhelming lion's share.  

Google's AI-assisted web search in my opinion inhibits one's ability to explore and learn via search. A query result as infopellet spat by a chatbot might (hopefully, not certainly) provide correct information, but it imparts no new understanding to the user. The user has a fact. They do not have the context that would have been provided by doing a search and evaluating the results themselves. Therefore the fact is of limited use. 

You recently launched two new tools at searchtweaks.com, Peabody Search and Super Edu Search Powered by Diffbot. What do they do?

As more and more slop infests the web in general, I believe it's increasingly important to use special syntax to carve out specific search spaces, restrict your searches to sites you know have some kind of verification/authority, or both. Peabody Search and Super Edu Search use both those principles. 

Peabody Search lets you restrict your search query by year or span of years and build your query from media in a particular state, the media sourced from the FCC database, list of NPR stations, and newspaper data sourced from Wikipedia. So instead of saying "I want to do a Google search for Troy Jackson,"  you can say "I want to do a search for Troy Jackson in a set of Maine media from 2018-2024." The 2018-2024 is when Jackson was president of the Maine senate – Peabody Search has a Wikipedia add-on that allows you to look up topics and find relevant date spans.  

One of the first search tools I made was Super Edu Search. It uses a dataset from the Department of Education to assist the user in developing very specific searches of .edu Web space. If you want to search public universities in Florida for "hurricane preparation," you can do that. If you want to search all Roman Catholic universities for the query "reproductive rights," you can do that. 

Super Edu Search Powered by Diffbot does the same but provides results from Diffbot's Web API instead of generating Google search queries. Diffbot recently released a Web API and I wanted to try it. 

SearchTweaks tools typically run via Google. Do you use other general search engines in your work and where do you find them particularly useful compared to Google?

I use Google because it's what people are most familiar with and because it has a robust date search offering. But I do not use it exclusively. My web site RSS Gizmos uses Bing in a couple of its tools, for example. 

There are other search engines I do use, like Mojeek. Both Mojeek and Brave have fairly inexpensive search APIs so I use them for my homemade alerts service, Calishat Snaps, and other personal search projects. Unfortunately I can't use their APIs for my free public stuff because that would get real expensive real quick.

You've been building and using search and research tools for a long time. What changes have LLMs brought to search, and how have you personally adapted?

I avoid AI results if at all possible. If I need to search Google I will usually append my searches with -aerwarwete -werawerwrew and that's often enough to turn the AI off. If I have quick reference searches I need to do, like find a Web site for an organization, I'll bypass Google and use MiniGladys. 

There are so many things to object to about the use of AI in search – the environmental costs, the misinformation, the abuse of intellectual property, etc – but what concerns me the most is the non-transparency of the results themselves. You ask Google's AI a question, you get an infopellet. How is that infopellet made? Where does the information come from? (Google has been better about providing source information, to their credit.) Is Google in agreement with any national government to exclude or specifically include data from/to that infopellet? Has anybody paid to include themselves in that pellet? 

… Despite all that I do feel that there is a place for AI in search, but it's assistive and complementary instead of baked-in and inescapable.

📍 Spyboy released CyberSuite, a Chrome extension “that brings a collection of web reconnaissance, security auditing, privacy, OSINT, and developer utilities together inside the browser.” It includes DNS and WHOIS lookups, email and username enumeration, and more. More info here.

📍 Andrew Deck of the Nieman Journalism Lab wrote about how fact-checkers are using Backstory, a Google tool currently in limited release. It “can automatically run checks to determine whether the image is AI-generated, whether it shows signs of manipulation, and where else the image has appeared throughout its lifetime on the internet.”

📍 Roman Huliak wrote, “‘Publicly Available’ doesn’t mean ‘Fair Game’: OSINT Automation and GDPR.”

📍 Jeremy B. Merrill wrote, “Using LLMs in Data Journalism Can Be Trustworthy, If These Five Elements Are In Your Methodology.”

📍 AI music generator Suno released a tool to check for its credentials on audio files. (FWIW I tested the files generated as part of last week’s audit; the detector flagged all four unedited Suno files, but caught only one of the edited tracks. — Alexios)

📍 German newspaper Die Tageszeitung published an article about OSINT. Dr. Manisha Ganguly, an investigations correspondent with The Guardian, was quoted in the story and also shared some of her thoughts on LinkedIn:

📍 Nico Dekens gave a talk at DEFCON, “OSINT Is Still a Thinking Game: Surviving AI Without Losing Tradecraft.” He published his slides and an overview of his key points on GitHub, and you can watch the talk here:

📍 The Israeli government is trying to influence how chatbots answer questions about Gaza and the Israel Defense Forces, according to POLITICO. The outlet reviewed newly filed US foreign agent disclosures naming the Hanover Institute, a website styled as a think tank. The disclosures show that more than a dozen of its articles were written by Piro, a US ad agency, as part of a $100,000 contract with Israel's government advertising agency. Piro subcontracts for French PR firm Havas Media, which handles the bulk of Israel's FARA-registered influence work in the US. The filings describe the work as educating the American public.

📍 Stanford political science professor Adam Bonica published a damning review of Democratic Party fundraising methods, especially with respect to elderly voters. Bonica wrote that spam text operations don’t “‘churn and burn’ through disposable donors, as the model is often described. Rather, they hook and squeeze.”

📍 C4ADS analyzed 19 months of flight data to link 75 flights to Russian state-aligned military operations in Mali, Burkina Faso, and Niger.

📍 Google placed programmatic ads for its own products on conspiracy theory websites that claim the 2020 US presidential election was stolen and that black seed oil can cure cancer, according to a NewsGuard analysis.

📍 Nitasha Tiku and Faiz Siddiqui spoke to a woman whose stepfather reportedly used an image taken when she was 11 to create 7,000 explicit deepfakes of her. The man, who went on to commit suicide, used Grok to generate the synthetic nudes.

📍 A small Australian consulting firm called Horizzen was impersonated as part of an alleged Chinese espionage operation that invited Western experts to apply for nonexistent jobs.

Want more studies on digital deception? Paid subscribers get access to our Academic Library with 75 categorized and summarized studies:

Please upgrade to access all of our content, including our how-to guides and Academic Library, and to our live monthly workshops.