In 1996, submitting a website to Yahoo meant convincing an actual human being that your business deserved to exist.
Nicole M. Radziwill worked as a systems administrator, programmer and project manager through that decade, and she remembers the ritual precisely. “I was a sysadmin for an ‘ecommerce shop’ in 1995 and 1996,” she said, “and when we would turn up websites for new clients, the highlight of our process was submitting the site to Yahoo. Yahoo was like the Yellow Pages, but only for websites. There was a form you would fill out, and you had to justify to the real people at Yahoo that this business you were submitting was legit and important enough to be in Yahoo’s main directory.”
Sometimes the humans said no.
“I remember one time submitting the website for a regional branch of the American Cancer Society and getting rejected because it ‘wasn’t significant enough,'” Radziwill said. “They recommended we contact the main ACS and have them link the site from their page… that they didn’t have yet.”
The assumption that you could search for everything didn’t exist yet
That’s the part people get wrong about the pre-Google web. The story isn’t that early engines were worse. It’s that they served a different Internet, one where directories carried real weight, crawling was still a craft, ranking was fragile and “search” hadn’t collapsed into a single dominant box.
People found pages through Yahoo-style directories, bookmarks, newsgroups, email signatures and links passed hand to hand. The web was small enough that human organization could still compete with machine indexing. Search engines existed. They were one option among several, not the front door.
Usenet got there first. Conceived in 1979, it handed users topic-based newsgroups and a culture of distributed discussion, which made online information feel communal well before the first major web engines showed up.
“It was the way to find out about websites that might interest you,” Radziwill said. “I got on Usenet in 1990 as a student in Durham, North Carolina… it was delightful, especially the alt.* and misc.* groups. People would put links in at the bottom of their posts, and awareness grew organically. And if you wanted safer or more reliable recommendations, you could limit yourself to moderated groups.”
Webrings ran on trust nobody thought to question
“Then a couple of years later, webrings popped up,” Radziwill said. “You’d get a block of code and put it on the bottom of your site’s HTML page, and it would embed a link to the next website related to this one. Someone else would manage the list of what could come next, so it was a great way to increase your exposure. The people putting together the lists for the webrings were generally pretty upstanding, so we didn’t even think about sabotage.”
Read that last sentence again. An open, unverified redirect chain, maintained by volunteers, and the threat model was nobody’s concern.
Webrings grew microcommunities the way the BBS era did, out of people with a shared niche interest who were sometimes neighbors in the physical world. Long before Reddit swallowed that function, a webring was one of the few ways to find sites about the thing you cared about.
The web wasn’t a retrieval machine yet. Services were built to let you browse topical categories rather than fire a query at a universal index, and a directory wasn’t a fallback. It was often the whole point.
Human curation worked better than it should have. Editors could spot quality, filter junk and impose order on a medium that was growing fast. But it couldn’t scale, and it got slower and more expensive as content outran the people classifying it. Once pages outnumbered editors, crawlers won by default.
AltaVista handed you 40,000 results and wished you luck
AltaVista, Lycos, Excite and HotBot attacked the scale problem with software instead of staff. They didn’t all work alike, but they shared an ambition: hoover up as much of the web as possible and let the algorithm sort it out.
AltaVista pushed hardest. It paired a fast crawler with indexing software that scaled, and it was fielding millions of HTTP requests per day shortly after launch. Searching stopped feeling like flipping through a catalog and started feeling like querying a machine.
Later versions supported natural language-style queries. For anyone still working out what the web even was, that came close to miraculous, even with results that were rough at the edges.
Mark Friend, director of the IT support firm Classroom365 Limited, worked as a systems operator in the late ’90s and doesn’t romanticize it.
“Most people have completely forgotten how chaotic it really was,” Friend said. “Back then, if you typed a question into AltaVista, the odds were stacked against you if you were looking for anything specific. You’d receive 40,000 results that would leave you just as confused as shouting into a crowded room.”
Working an AltaVista results page was a skill you built over months. But AltaVista fixed one expectation permanently: search should be instant. Once people felt a huge index swept in seconds, partial and sluggish systems were finished. AltaVista lost the war and still wrote the rules of engagement.
Lycos and Excite wanted to be destinations, not tools
Both were search brands and portals at once, with news, email, sports, weather and finance stacked around the query box, all of it engineered to keep you on the page instead of sending you off to results.
Nobody agreed on what the category even was. Some engines chased breadth, some chased speed, some blended editorial channels with automated results in a way that reads as bizarre now. “Search engine” covered directories, crawlers and portals with a search field bolted on as an afterthought.
HotBot came off as the serious one. It landed just as users started noticing that quality depended on index size and ranking sophistication together, and its scalable backend came from Inktomi. A fast crawl wasn’t enough. People wanted results that were relevant, current and not obviously gamed.
There was the weak point. Lean too hard on on-page text signals and anyone can stuff a page with word bloat to climb. Marketing beats quality. The early web turned into a manipulation lab before anyone was saying “search engine optimization.”
Ask Jeeves promised conversation the technology couldn’t deliver
Ask Jeeves felt futuristic to Friend because it asked you to type a question instead of thinking in keywords, which gave the impression of a knowledgeable assistant running the web on your behalf. The resemblance to today’s chatbots is easy to overstate, and worth resisting: Ask Jeeves offered nothing close to an LLM-based conversational interface.
Natural language is hard. Users didn’t ask clean questions, and the systems underneath couldn’t infer intent reliably at scale. Ask Jeeves is memorable for claiming to understand people better than the technology of its moment could support.
Hiding white text on a white background was just a tactic
Optimization existed. The professionalized version, with departments of experts chasing Google’s next mysterious pivot, did not. Early engines bent to obvious signals: keyword repetition, metadata, submission tactics.
“People didn’t talk about ranking back then,” Friend recalled. “They submitted their URL and waited patiently to see if it would show up in a search. Everyone took the Meta keyword tag seriously and the practice of utilizing ‘white text on a white background’ to hide hidden keywords was a legitimate tactic that webmasters would use to include keywords in their website for search engine crawlers. There was no fear of penalties, and you built a website, crossed your fingers, and hoped for the best.”
By the late ’90s engines had named that trick spamdexing and started punishing it. Ranking was simpler then, and more fragile. The rules hadn’t hardened, and the feedback loop between publishers and engines stayed small enough to test and read. Once search became the main gateway to the web, that ended and the relationship turned adversarial.
“The hard truth that most people do not want to face is that scraping and ranking turned the Internet into a manufacturing facility of content,” Friend said. “Websites went from being written for people to being written primarily for search engine crawlers.”
Google’s real invention was the logic of ranking
It didn’t discover search. PageRank treated links as signals of authority, which made relevance partly a question of how the rest of the web pointed at a page. Results felt more trustworthy, and crude manipulation got harder. Not impossible.
The interface did as much work as the algorithm. A stripped-down page that got out of the way told users that search was a utility rather than a portal amusement park, and the search box became the front door to everything.
The era it replaced was slower and messier. It also held more competing theories about how information should be found, sorted and judged. Some systems trusted people, some trusted crawlers, some trusted directories, some trusted a question typed in plain English.
“The Internet contained an inherent level of clutter and chaos,” Friend remembered. “But at the same time, it was a much more humanized place before. It’s why many of us feel nostalgic for that time. The Internet had an artisanal feel to it. It was made by real people using text editors such as Notepad and Dreamweaver. You might start at one destination and find yourself at a fan site of an obscure band, and then end up at a forum discussing vintage synthesizers, and ultimately end up on a NASA page.”
Those engines weren’t failed drafts of Google. They were serious answers to a problem the web had made urgent without making solvable, and search was never fated to look the way it does now. Google won by putting technical rank, usability and scale together at the exact moment everyone else was ready to give up the old order.
