How Search Should Work

Everyone complains about Google Search. I've done my share of that too. But complaining doesn't build anything.

This is a blueprint. A concrete system for how internet search should work, with real code and an interactive prototype you can try yourself. I'm not claiming this is the only way to do it. I'm claiming it's better than what we have. Look at it and decide for yourself.

The Reactive Trap

Google has spent its entire existence fighting spam. Every algorithm update, every manual action, every new signal they add to their ranking system is a response to abuse that already happened. They're playing defense. Always.

The world's largest search engine, backed by a two-trillion-dollar company with some of the smartest engineers on the planet, spends most of its energy reacting to people trying to game the system. This is the design flaw. Not a bug in the implementation. A flaw in the architecture itself.

When anyone can create a website and have it indexed automatically, you inherit every problem that comes with an open, unverified system. Spam, manipulation, content farms, AI slop. Google's answer has been to build increasingly complex filters to catch bad actors after they've already polluted the index. Those filters inevitably catch good actors too.

There's a better starting point.

Step 1: Registration

Before your site appears in search results, you register it. With your real name. Not a company name hiding behind a P.O. box. A person who is accountable for the content on that site.

This is the single biggest structural change you could make to search. I almost never hear anyone talk about it.

Here's what registration looks like:

• Your legal name (verified)
• Your website URL
• A brief description of what your site covers
• Contact information

That's it. Simple form. Five minutes.

"But there are billions of websites!" Sure. How many of those are legitimate sites created by real people who want to be found in search? A fraction. The overwhelming majority of the web is parked domains, spam, abandoned pages, and machine-generated garbage. You don't need to index all of that. You need to index the sites that have something real to offer.

"But what about anonymity?" This is a real tradeoff, and I want to be honest about it. If you're running a site for whistleblowers, political dissent, or sharing personal stories you can't attach your name to, this system isn't built for that. Other channels (social media, direct links, newsletters, dedicated platforms) serve those needs. What this system is built for is the millions of normal websites publishing normal content for normal people searching for answers. For that use case, accountability is a feature, not a cost.

Now think about what happens to spam. If you're running a network of auto-generated affiliate sites, are you going to put your name on each one? Submit yourself to scrutiny? No. You move to wherever there's no friction. And that's the point. You don't need an algorithm to filter spam if spam never enters the index.

Step 2: Declare Your Lane

This is where it gets interesting.

When you register, you declare your expertise. What topics does your site cover? What's your lane?

Forbes declares "Business News and Financial Analysis." SoundGear Reviews declares "Personal Audio Equipment." A solo blogger in Chiang Mai declares "Budget Travel in Southeast Asia."

These declarations are public. Searchable. And they have consequences.

If your site publishes content within your declared expertise, the algorithm treats it with full weight. If you publish outside your lane, the content still gets indexed, but it competes at a disadvantage. Not as a penalty. As a structural reality: you haven't built the depth, the track record, or the trust in that space.

This solves the Forbes Advisor problem. Forbes is a legitimate publication with real journalists. Their business reporting is excellent. But they also publish "Best CBD Gummies" and "Best Mattresses" and "Best Web Hosting" because those keywords are profitable. Under the current system, Google's brand signals boost this content regardless of whether Forbes has any actual expertise in mattresses. Under this system, that content competes against sites that have been testing mattresses for years. And it loses. Not as punishment. Because the people who actually test mattresses are better at writing about mattresses. The system just reflects that.

Declarations aren't permanent. Sites evolve, coverage shifts. You update your declaration. But there's a history. The algorithm sees that you've been covering audio equipment for six years, or that you added "cryptocurrency" to your topics last month. History matters.

This changes the incentive structure entirely. Instead of optimizing for what Google's algorithm rewards this quarter, you optimize for something stable: being good at the thing you say you do.

Step 3: The Algorithm

Here's the ranking system. I've written it as code because code is precise, and there's no room for hand-waving.

function rankResults(query, index) {
  // Only registered, verified sites enter the ranking
  const candidates = index
    .filter(site => site.verified === true)
    .filter(site => hasRelevantContent(site, query));

  return candidates
    .map(site => ({
      ...site,
      score: calculateScore(site, query)
    }))
    .sort((a, b) => b.score - a.score);
}


function calculateScore(site, query) {
  const expertise    = scoreExpertise(site, query);
  const quality      = scoreContentQuality(site, query);
  const satisfaction = scoreUserSatisfaction(site, query);
  const consistency  = scoreConsistency(site, query);
  const freshness    = scoreFreshness(site, query);

  // Five signals. Weighted by importance.
  return (
    expertise    * 0.30 +
    quality      * 0.30 +
    satisfaction * 0.25 +
    consistency  * 0.10 +
    freshness    * 0.05
  );
}

Five components. Each one scored between 0 and 1.

Let me break them down.

Expertise Match (30% of the score)

The most important signal, shared with content quality. Does the query fall within what this site actually does?

function scoreExpertise(site, query) {
  const declaredTopics = site.declaredTopics;
  const queryTopic = classifyQuery(query);

  // Direct match: query falls squarely within declared expertise
  if (isDirectMatch(queryTopic, declaredTopics)) {
    const years = yearsSince(site.topicHistory[queryTopic].startDate);
    const depth = Math.min(years / 5, 1.0); // Caps at 5 years
    return 0.7 + (depth * 0.3);             // Range: 0.7 – 1.0
  }

  // Adjacent: related to declared expertise but not core
  if (isAdjacentMatch(queryTopic, declaredTopics)) {
    return 0.4;
  }

  // Outside declared expertise entirely
  return 0.1;
}

A headphone review site that declared "Personal Audio" five years ago scores 1.0 for a headphone query. Forbes, which declared "Business and Finance," scores 0.1. Not because Forbes is bad. Because this isn't what they do.

The 0.1 floor (not zero) means off-topic content can still appear if it's exceptional on every other signal. But it has to clear a high bar. An article that's outside your lane needs to be great on quality, satisfaction, and consistency to surface. Most won't be.

Content Quality (30% of the score)

This is where AI earns its place. Not generating content. Evaluating it.

function scoreContentQuality(site, query) {
  const page = findMostRelevantPage(site, query);

  // AI assessment of the actual content
  const depth       = assessDepth(page);
  // Does it cover the topic thoroughly?
  // Surface-level listicle vs comprehensive analysis.

  const originality = assessOriginality(page);
  // Is this unique analysis, or repackaged from other sources?
  // Original testing data, personal observations, novel angles.

  const evidence    = assessEvidence(page);
  // Does it cite sources? Show data? Describe methodology?
  // "Great noise cancellation" vs measurement graphs
  // and A/B comparisons against 5 competing models.

  const experience  = assessFirstHand(page);
  // Signs of actual hands-on experience.
  // Original photos. Specific observations about comfort
  // after 8-hour sessions. Wear patterns after 6 months.
  // Details you can only know if you held the thing.

  return (
    depth       * 0.25 +
    originality * 0.30 +
    evidence    * 0.25 +
    experience  * 0.20
  );
}

A page that says "The Sony WH-1000XM6 has great noise cancellation" with nothing backing it up scores low on depth and evidence. A page with measurement graphs, side-by-side comparisons, photos of the testing setup, and specific notes about how the headband pressure changes after week three scores high.

Can this be gamed? Eventually, yes. Someone will use AI to generate convincing fake testing evidence. But remember: their name is on the registration. Their reputation is the collateral. The incentive to fake evidence is much lower when the cost of getting caught is personal and permanent.

People like HouseFresh and Retro Dodo, the ones who actually care, will never need to fake anything. They already have the real thing. This system just makes sure the real thing wins.

User Satisfaction (25% of the score)

This is the signal Google already has but buries under a hundred other factors. Did the person find what they were looking for? That's the whole point of a search engine.

function scoreUserSatisfaction(site, query) {
  // Did the user come back to search the same thing again?
  // Lower is better: they found their answer.
  const searchReturn = measureSearchReturn(site, query);

  // Time on page, scroll depth, interaction.
  // Normalized against content length. Reading a 5000-word review
  // should take longer than scanning a 500-word summary.
  const engagement = measureEngagement(site, query);

  // Do users come back to this site on their own?
  // Direct visits, bookmarks, return visitors.
  // Slow-building signal. New sites start at 0.5.
  const loyalty = measureLoyalty(site);

  return (
    (1 - searchReturn) * 0.45 +
    engagement         * 0.30 +
    loyalty            * 0.25
  );
}

If someone clicks your result and doesn't come back to search, you answered their question. If they bounce back in three seconds, you didn't. This isn't new technology. Google measures all of this today. The difference is that in this system, it's 25% of the score, not a minor signal diluted by a thousand others.

The loyalty component is interesting. Sites where people type the URL directly, where they have bookmarks, where the same visitors return week after week: these are sites people trust. That's hard to manufacture.

Consistency (10% of the score)

function scoreConsistency(site, query) {
  const queryTopic = classifyQuery(query);
  const articlesInTopic = countArticles(site, queryTopic);
  const totalArticles = countArticles(site);
  const topicFocus = articlesInTopic / totalArticles;

  // Depth: reward a body of work, not a single article
  const depthScore = Math.min(articlesInTopic / 20, 1.0);

  // Regularity: consistent publishing vs sporadic bursts
  const regularity = measurePublishingCadence(site, queryTopic);

  return (
    depthScore * 0.40 +
    topicFocus * 0.30 +
    regularity * 0.30
  );
}

A site with 50 articles about headphones, published regularly over three years, scores higher than a site with one headphone article mixed in with content about cryptocurrency, cooking, and pet insurance. Consistency signals commitment to a topic.

Freshness (5% of the score)

function scoreFreshness(site, query) {
  const page = findMostRelevantPage(site, query);
  const daysSinceUpdate = daysBetween(page.lastUpdated, today());

  // Topic sensitivity: tech reviews need to be current.
  // History articles don't.
  const decayRate = getTopicDecayRate(classifyQuery(query));

  return Math.exp(-daysSinceUpdate * decayRate);
}

Freshness has a low weight on purpose. A well-written headphone review from eight months ago is still valuable. The decay rate adjusts per topic: product reviews decay faster than educational content, news decays fastest of all.

What It Looks Like

Theory is one thing. Here's what it looks like for someone actually searching.

Scroll through the indie results: every publisher has a name, a track record, and a satisfaction score. Then click "Google Today" and look at what the same query returns. Four ads. An AI overview pulling from the same sites that will appear in the organic results anyway. A "People also ask" box. A Reddit thread. Then, eventually, some actual results.

Count the steps between the search bar and the first piece of content written by someone who actually tested headphones. On the Open Search side, it's zero. On Google, it's at least six.

What About Gaming?

People will try to game this system. They always do. That's not a reason to dismiss it. It's a reason to look at the attack surface and compare it to what we have now.

Under the current Google system, gaming is essentially anonymous. You can create hundreds of websites under different names, build link networks, use AI to generate content at scale, and there's no identity attached to any of it. If one site gets penalized, you spin up another. The cost of getting caught is near zero.

Under this system, every manipulation traces back to a registered person. Build a fake review site? Your name is on it. Create a network of sites to cross-link? All registered to you. Generate fake user satisfaction signals? When the pattern is detected, it's linked to your identity, not to a disposable domain.

The people who try to game search engines are rational actors. When the cost of getting caught goes from "lose a domain" to "lose your reputation," the math changes.

Meanwhile, the people who have never needed to game anything, the ones who just make good stuff, they're unaffected. They register, declare their lane, and keep doing what they've always done. The system rewards them instead of punishing them for not playing the SEO game.

The Tradeoffs

I said I'd be honest, so here's what this system costs.

Anonymity. Some valuable voices on the internet are anonymous by necessity. This system doesn't serve them for search specifically. The web is bigger than search, and distribution channels exist that don't require identity. But there's a real cost here, and I'm not pretending otherwise.

Cold start. A brand new site with one article and a fresh registration is hard to rank. That's a real disadvantage for newcomers. The system accounts for this: new sites get a neutral baseline for user satisfaction (0.5) and loyalty. Expertise match works from day one. If the content quality is high, the site can surface. But it won't outrank a site with five years of track record overnight. That's by design. Depth should matter.

Scale. Reviewing registrations, maintaining identity verification, handling disputes. This costs money and people. A search engine running this system would need a trust and safety team. Google has 182,000 employees. They could staff this tomorrow if they chose to.

Who loses. Big media companies publishing SEO content outside their expertise. Content farms with real names but no real knowledge. SEO consultants whose value proposition is gaming the algorithm. Affiliate networks that exist to redirect traffic, not to inform. These entities would rank lower, possibly much lower. Some would disappear from search results entirely.

I consider this a feature.

I spent more time on this than I probably should have. Not because search is a personal passion, but because I got tired of waiting for someone to propose something concrete. Every conversation about Google goes one of two ways: "Google is terrible" or "well, search is really, really complicated." Neither moves anything forward.

So here it is. Identity verification, expertise declarations, quality assessment, user satisfaction. None of this is science fiction. A team of fifty engineers could build a working prototype in six months.

Google won't build it. The current system, broken as it is, generates $307 billion a year in revenue. A system where Forbes Advisor doesn't rank for "best mattress" is a system where Forbes doesn't need to buy Google Ads for "best mattress." Follow the money.

Someone else should build it. The algorithm is on this page. The interface is on this page. The blueprint is here.

Use it.