On April 6, 2021, Yahoo Answers (YA) announced it would be shuttering its doors and taking 15 years of content offline today, May 4, 2021. It’s a clumsy end to a promising beginning for the infamous and storied platform.
Often the target of memes, ridicule and trolling, for many years YA became a lightning rod for what was wrong with the internet. Internet denizens have come out in full force to share their favorite one-liners and memories from a long and illustrious history.
A look back at the rise and gradual fall of YA is invariably linked to the mismanagement of Yahoo as a corporation, just as much as the mismanagement of the platform itself. It’s also another failed example in the Q&A space that was already littered with failed companies, poor business models and wayward investments.
But before we dive into what went so horribly wrong in the questions and answers space, let’s take a look back to the early days of YA. There are many clues there as to what was to come.
The origin of Yahoo Answers
Yahoo was founded in 1994 by founders Jerry Yang and David Filo. As one of the early pioneers of the Internet, Yahoo provided the first “directory” of all websites on the internet and had the endearingly innocent name of “Jerry and David’s Guide to the World Wide Web.” Early Yahoo more resembled a really well-organized Windows File Explorer for websites than today’s page-rank based search engines.
The directory was managed by humans with a hierarchy created and curated by humans. To be placed in the directory, website owners would submit their domain names to a group of 20 “classifiers” that all sat in a room together and attempted to provide consistency to Yahoo’s ontology. It was a model that was painstakingly curated and replete with value while also being a model that was infinitely unscalable as the internet grew. Surprisingly, Yahoo Directory would last a full 20 years before finally being shut down in 2014.
Yahoo grew rapidly over its first 10 years and quickly became a web portal offering diversified services. One of those services, Yahoo Answers, would officially launch in 2005 with immediate traction. Users flocked to the new Q&A service from the early days of its beta and quickly made it the second largest reference website in the world behind only Wikipedia. Yahoo heavily promoted the service in its search engine, and the ease of use and sense of community contributed to strong early growth numbers. At one point in June of 2006, YA could count .044% of all internet users as a monthly visitor. That’s an astounding number for a website that had its beta launch just one year prior.
The premise of YA, like similar websites that came before and after it, was to assume that a group of people could do a better job finding information on the internet than a single individual or the best of search engines. Additionally, the baseline principle of YA was that most knowledge was stuck locked away in peoples’ heads instead of on the internet and needed a way to come out. Conceptually, human knowledge was, and still is, more of an iceberg than a container ship.
What contributed to Yahoo Answers’ early growth numbers?
The popularity of YA skyrocketed for a number of reasons. Its simplicity was an asset to start. YA provided their users a place to ask more complex questions, questions that required personal or professional experiences, human empathy, and the ability to apply reasoning to a challenging problem. These are all things computers and search engines were unable to do at the time, and, for the most part, still can’t.
In the early days, Yahoo threw a lot of weight behind marketing and celebrity influencers, bringing on guests like Stephen Hawking and Bono to ask questions. Hawking’s question “How can the human race survive the next 100 years?” garnered over 25,000 answers.
Community was also a big reason why people flocked to YA. It was a place where people could ask questions and get advice and help from other people and their experiences. This concept has a name — social search.
To understand the success of social search websites like YA and Google Answers, it’s important to contextualize the time period when these sites were garnering such strong user traction. Search engines of 2005 were nowhere near the juggernauts of information and link correlation that they are today. While it was possible to find answers to most fact-based questions at the time, it was also very easy to confuse and confound a search engine with poorly formed queries. YA gave a home and, more importantly, a voice to people struggling to find help and advice from a burgeoning online world.
Early media coverage
Early media coverage of YA was mixed. Reporters tended to be impressed with the early user growth numbers but skeptical of the overall ability of social search solutions to deliver true value.
Trust is an inherent problem in any social search platform, and YA was no exception. The irony of YA was that social search was intended to establish trust by adding human eyes and reasoning to the user’s experience. The more verification points (answers from multiple users in this case), the more confident a user would feel in the aggregate answer.
Calling on the collective knowledge of the internet community to answer questions posted by other members of the web community seemed like a great idea. There are still many similar websites that rely upon this same principle including online forums like Quora and Reddit, and to some extent next-generation knowledge marketplace platforms like Sage (which, full disclosure, is the company founded by this author).
But adding more opinions comes with its own set of moderation problems. And this is where things started to break down for YA. Their moderation system was built and managed by a customer success team. Each reported violation was looked at by a pair of human eyes to make moderation decisions. Not only does this not scale, but YA also reported a 10% error rate from the customer success team’s decisions.
The most proven and effective way to approach large-scale moderation problems is with community-driven moderation. This is a critical component to any social search experience, and without a disciplined approach to your moderation strategy, quality of user experience will continue to degrade.
Bumps in the road
The first signs of trouble aren’t always obvious: allowing questions to be posed as forms of commentary or questions that aren’t really questions at all. These and other examples began popping up on YA with regularity. The anonymity of the website was becoming a problem.
Creating a barrier to entry is an integral step to a strong knowledge exchange experience, but Yahoo went the opposite direction with its Q&A service: It made it easy to ask anonymous questions. It made moderation a confusing and obtuse corporate process. And the results came quickly and furiously in the form of technical and user experience debt as the abusers, trolls and others realized how easily they could turn the platform from a knowledge source to a source of entertainment.
Questions like “how was babby formed?” and “how do I unbake a cake?” while amusing sources of entertainment, did nothing to promote the platform as a true source of knowledge. The lack of moderation efforts was especially confounding when looking at Yahoo’s plans for the platform. At the time in late 2006, YA was a clear part of Yahoo’s strategic plan. It intended to combine the concepts of search indexes plus social search to give themselves a leg up on their dwindling search market share:
“The integration of Yahoo! Answers with Web search demonstrates the Yahoo! Search vision to enrich people’s lives by enabling them to find, use, share, and expand all human knowledge,” said Eckart Walther, vice president of products at Yahoo! Search. “Yahoo! is providing a better search experience by making it easier for people to tap into the collective knowledge of others online for everyday questions.”
It was clear Yahoo executives had big plans for YA. However, combining poor moderation with high visibility was not a winning formula. As more and more answers content became visible in the Yahoo traditional search results, the reputation of Yahoo Answers as a platform began to wane.
Software efforts to improve the platform
In the summer of 2007, it was becoming apparent that Yahoo needed to find a better solution to the problem of moderation. Their customer care terms were overwhelmed with abuse reports, and their efficacy began to fail at keeping YA content quality high. The current reporting mechanisms weren’t hiding abusive and offensive content quickly enough, and user complaints were skyrocketing.
The YA business unit needed to tackle this problem head on. They assembled a team consisting of Director of Engineering Ori Zaltman; Product Manager Quy Le; Anirudh Koul, who led troll hunting and optimizing of the model; and Micah Alpern, lead user experience designer.
The team would also rely heavily on an already built reputation platform team within YA. Yvonne French was the product manager for the reputation platform, and “Building Web Reputation Systems” coauthor Randy Farmer was the platform’s primary designer and advised on reputation model and system deployment.
They had two primary goals:
- Remove abusive content from the website faster.
- Reduce error rate on content moderation.
False positives hovered around 10%. The average amount of time offensive content stayed onsite before being removed was 18 hours. Their goals were to reduce this time to one hour while maintaining a low 10% false positive rate.
They would do this by removing customer care from the moderation workflow for all but the most egregious situations. A new concept of “hiding” the abusive content was also introduced as this removed external approval mechanisms that would slow down the entire process. When the new algorithms determined content was abusive, it would be hidden immediately while giving the content’s creator the ability to appeal the process.
If the appeal was granted, reputation scores of the reporters of abuse would go down while the reputation of the content’s creator would go up. Conversely, if the appeal was denied, the reporters’ reputations would rise while the creator’s reputation took another significant hit. As the system improved, as the reporters either gained more or less credibility depending on their accuracy and appeal decisions, the entire process would become more and more efficient.
Trusted reporters would be able to remove content almost immediately with fewer additional reports required to verify the action. Reporters with low reputation would be unable to affect content on the website. And content creators would be controlled through a combination of poor reputation, limited website access, or outright bans in some cases.
The new reputation system was a rousing success by all measures as the goal of one-hour removal was far surpassed and error rates were a fraction of the customer care team’s. Here’s the in-depth case study on how the new system worked.
Trust, identity and reputation
So why did YA ultimately fail? It was never because of lack of interest. In fact, the popularity of YA continued to soar through the late 2000s and growth continued to exceed all expectations. In 2010, it hit their 1 billionth Q&A transaction on the website. At that point, YA was seeing 823,966 Q&As per day. A quick math equation gives us 34,331 per hour or 572 per minute — equating to 10 Q&As every second. Those are wild numbers and truly show just how much value the internet community was getting from the platform.
But YA was missing key ingredients that a knowledge marketplace platform needs to survive.
There is a feeling of trust when getting advice from other people. The more you know the person, the more you trust in the advice and counsel you receive. Posting your answer to an online web forum where anonymity is a protected asset can break down that trust barrier beyond repair. Yes, it was possible to get great advice from someone named “Semi-Mad Scientist,” but it was also possible to get awful advice from a user named “onlyGreatAdvice.”
YA’s solution for this was to gamify the system. The only real measure within YA was the ability to accept someone’s answer and then view what percentage of answers from that user had been accepted. A fine measure but one that requires several other triangulation points before it begins to tackle the problem of establishing trust.
One of the things you did not get with YA was any concept of identity. While you did get answers from other human beings, you did not know who those human beings were, what their credentials were, or how their experiences prepared them to give you credible advice. YA gave all users the ability to both ask and answer questions anonymously or with pseudonyms.
In the successful versions of YA you had people on the other end of your post who thought about your question, applied their own human reasoning, their powers of empathy, and their ability to put all of this together into a single and comprehensive post.
This was the value of Yahoo Answers: to know you were getting an answer or answers from other people who had been in your shoes and walked the path you were now walking. That gave you comfort, solace and confidence that the information you were getting was dependable and trustworthy.
Unfortunately, identity is a key concept to any exchange of trustworthy and reliable information. Without it, you can’t verify someone has the background knowledge to help you. You can infer it from their answers to some extent, but kickstarting that trust from a point of zero-trust with a single answer just doesn’t work. You can go back and look at their previous answers to gain more confidence, but that requires time and manual effort.
Quora saw the failures in identity from platforms like YA and Ask Jeeves and required that people register with their real names because of this. That helped to establish some identity as you could Google a person’s LinkedIn profile or social media and glean more information to help understand if they were a reputable source. But again, time is precious and adding burden to the consumers of your system eventually fails when fast, reliable knowledge exchange is the end goal.
Enter the world of reputation. When YA’s engineering team tackled the problem of reputation in 2007, they made a critical mistake in the interest of maintaining anonymity and entertainment value of the platform. They made the entire system operate and measure user activity behind the scenes. While YA was measuring each and every interaction a user had with the website, they did not expose critical metrics from those measurements that were invaluable to showing transactional experience on the platform.
How do humans typically build interpersonal trust in each other? They do this through a series of interactions or transactions. Those transactions can be verbal, visual or crowdsourced.The more I learn about a person, the more information I have to glean the types of value I can get from a relationship.
Verbal transactions are simply conversations. The more I talk to someone, the more I understand their areas of interest, the topics they are an expert in, and their personal and professional backgrounds. Each conversation is a transaction I subconsciously use to evaluate their credentials.
Visual transactions occur each and every day in our lives. If I need help planting and maintaining my garden and my next door neighbor has a gorgeous garden that seems to bloom multiple times during the course of the year, I can trust they know what they are talking about. I can visually verify that she is an expert in gardening. When the time comes that I want to get my own garden going, I’m going to go ask her for advice on what to grow, how to enrich my soil properly, which plants will thrive in which parts of my yard, etc.
Crowdsourcing knowledge is the final transactional model and the one used by most online platforms today to varying effectiveness. YA had a very simple model where users accumulated points mostly by being active on the platform. Quora has a different model that primarily revolves around content views. The more someone’s content is viewed, the more likely it is that the community finds the content helpful, the more likely Quora is to show the answers more prevalently, and the more obvious it is that their users are experts in the topic areas they choose to write about.
Another way to crowdsource is to measure every transaction that you can to effectively evaluate whether what someone says they know is actually what they truly know. Expertise isn’t measured by one good answer, but by the accumulation of proof from a thousand answers given.
Putting it all together
In order to create a comprehensive knowledge exchange platform, I believe you need to put all of the key components together. You must establish and foster trust amongst your users. You need to place identity front and center. Most importantly, you must create a comprehensive reputation system that measures all transactions on the platform to bring a virtual credentialing model to the forefront.
Once you’ve established a reliable online system to show user credentials that are verified both on- and off-platform, you can build your knowledge marketplace — but not before.
New platforms are starting to take shape to tackle this problem. (Sage is one of those that combines the ideas of trust, identity, and reputation with modern artificial intelligence and machine learning concepts to create a better knowledge exchange marketplace.) If you can quickly route a person’s question to the exact expert most likely to be able to answer their question, you’ve removed most of the friction inherent in today’s iterations of Q&A.
Efforts to save Yahoo Answers content
There are multiple efforts currently ongoing to save the content from Yahoo Answers. The Internet Archive team has started to index Yahoo Answers many times a day to try to pull down as much of the content as possible into their archive. In addition, a group of hackers known as the Archive Team has begun their own effort to pull down all the Yahoo Answers content through their web scraping scripts.
Yahoo Answers experienced a lot of their problems because of their inability to weed out the trolls and the poor content that ended up rampant throughout the entire platform. Misspellings were everywhere. Short questions that made no sense or were written as jokes or forms of entertainment dominated the front pages of categories. Moderation capabilities were created early on to help address these problems but were not continuously developed and improved.
Regular users, and even power users, were no longer able to effectively weed through the noise and allow the truly useful content to shine. Ultimately this is what ended up being the downfall of Yahoo Answers. It was never really clear whether Yahoo as a corporation wanted YA to be an entertainment platform, a knowledge sharing platform, or something in between. Its own users never figured this out, either.