How Googleranks a page
Walk through all eight on-page search steps, from crawling and HTML parsing to inverted indexes and multi-field scoring. Edit four sample pages right in your browser, tweak field weights, or type your own queries to see every table and search rank update live.
8 steps in total: the first five build the index, the last three answer a search. Everything runs locally in your browser.
Nothing you type or paste is uploaded. This simulator covers on-page text understanding and scoring only, not backlinks or domain authority.
Step 1 · Crawl: Discover URLs and fetch raw HTML
Crawlers discover pages through internal links and XML sitemaps. Before downloading anything, Googlebot checks robots.txt permissions, fetches the raw HTML response, and reads <meta name="robots"> tags such as noindex.
Step 1 decides whether a page can enter search at all. A leftover staging disallow rule, stray noindex tag, or 403/5xx server error blocks a page no matter how good its copy is.
How real Google compares
In production, Google also queues JavaScript-heavy pages for its Web Rendering Service (WRS), essentially a headless Chrome browser that runs your scripts before the page is parsed. Putting core copy straight into the initial server-rendered HTML gets pages indexed faster and more reliably.
Try this yourself
Pick any sample page in the crawl queue and toggle "Block via robots.txt" or "Mark noindex": it drops out of the live ranking at the top of the page and out of the Step 8 results. You can also click "Edit raw HTML" to rewrite tags directly.
<!DOCTYPE html>
<html lang="en">
<head>
<title>Complete On-Page Google SEO Guide: From Crawling to Ranking</title>
<meta name="description" content="Learn how Google crawls, indexes, and scores web pages. Master title tags, H1 headings, keyword relevance, and BM25 scoring principles." />
<style>.hero { color: #0f2942; font-size: 32px; }</style>
</head>
<body>
<nav>Home | Solutions | Blog | Pricing | Contact</nav>
<main>
<h1>Complete On-Page Google SEO Guide</h1>
<h2>How to Optimize Title Tags and H1 Headings</h2>
<p>Google SEO starts with clear HTML semantics. Placing your core keyword naturally in the Title tag and H1 heading helps search engines identify your page topic during indexing and scoring.</p>
<h2>Content Relevance and Natural Keyword Layout</h2>
<p>Modern search ranking evaluates topical relevance, inverse document frequency, and page length normalization rather than blind keyword repetition. Write focused guides that match user search intent.</p>
</main>
<footer>© 2026 Example SEO Lab · Privacy · Terms</footer>
<script>console.log('analytics init');</script>
</body>
</html>Clean titles vs body stuffing
For “google seo”, SEO Guide, with the query in its Title and H1, scores 5.51: about 5.5× the keyword-stuffed Stuffed Blog (1.01). Title weighting, diminishing returns on repeated words, and length penalties all push the focused page ahead.
5.51
SEO Guide (clean structure) · 2 body hits · Rank #2
Target keywords sit naturally in Title, H1, and H2 with tight, focused copy.
1.01
Stuffed Blog (counter-example) · 18 body hits · Rank #3
Repeats keywords over and over inside rambling paragraphs while Title and H1 miss the topic.
For “google seo”, the fix moves it from #3 to #1.
Turning these 8 steps into daily on-page SEO
Almost every routine on-page SEO check on a Shopify store or company site maps directly back to one of the eight steps above.
Step 1 · Crawl & indexCheck crawlability and indexation before rewriting copy
Site launches, theme changes, and staging pushes routinely leave behind robots.txt blocks or stray meta noindex tags. If a URL never enters the inverted index, better copy cannot save it.
How to check: Run URL Inspection in Google Search Console to verify live crawl status, rendered HTML, and the canonical URL Google chose.
Steps 2 & 7 · Title tagPut the core topic early in your Title tag and keep it concise
The Title tag carries the strongest on-page topic weight and acts as the clickable blue headline in Google. Putting the main subject before your brand name clarifies relevance and prevents awkward mid-word truncation.
How to check: Aim for roughly 50–60 characters (~580 px) and give every indexable page its own distinct Title.
Steps 2 & 7 · HeadingsKeep one focused H1 and map subtopics with H2–H6
A clear H1 backs up your Title’s topic, while specific H2 and H3 subheadings answer subtopics and buyer questions, earning extra heading weight and full-query matches.
How to check: Save H1 for the main page headline (never a navbar logo) and keep heading levels in logical order instead of styling text with random tags.
Steps 4 & 7 · FocusSkip keyword stuffing and cut off-topic filler
BM25 saturation caps gains from repeating one word quickly, while length normalization penalizes posts padded with unrelated tangents. Focused, direct answers outscore stuffed copy every time.
How to check: Read your opening paragraph and subheadings against the target query; cut rambling intros that dilute the main subject.
Steps 1 & 2 · Clean HTMLKeep main copy in initial HTML and trim template clutter
Bloated ecommerce themes packed with hidden menus, repeated promo blocks, or text injected only after client-side scripts run dilute main content and slow down rendering queues.
How to check: Open "View Page Source" in your browser and search for your main product copy to confirm it exists right in the raw HTML.
Step 8 · SERP snippetInclude search terms in Meta Description to reduce rewrites
Meta Description does not add ranking points directly, but when it naturally echoes the user’s query and states clear value, Google is far more likely to display it as written and lift CTR.
How to check: Write 130–155 characters specifically for each page’s target query rather than letting Google clip your opening paragraph.
What changes in AI search
AI answer features in search, such as Google’s AI Overviews and AI Mode, and AI assistants that search the web usually retrieve pages and passages first, then generate an answer. The retrieval half still depends on a page being crawlable, indexable and matchable, which is what Steps 1–6 showed. What differs is mostly what happens after retrieval.
| Stage | Classic search (what this simulator shows) | AI search | Verdict |
|---|---|---|---|
| CrawlStep 1 | The crawler checks robots.txt, then fetches the HTML the server returns; Googlebot also renders JavaScript. | Google’s AI features use the same Google Search crawling and index. Crawlers behind other AI assistants may not run JavaScript, so keep core content in the served HTML. robots.txt rules can allow or block each crawler separately. | Partly changed |
| ExtractStep 2 | Strip navigation and footers, then split out Title, H1, H2–H6, Body and Meta Description. | Clean main content still comes first. The difference is that content is often split into passages (chunks) that are retrieved and cited one by one, so each section should make sense on its own. | Partly changed |
| Tokenize & understandStep 3 | Split text into words, normalize case and word forms, and drop stop words. | Keyword matching is still used, but retrieval also uses semantic matching, comparing meaning as vectors (embeddings), so synonyms and paraphrases can match. | Changed |
| IndexSteps 4–5 | Build forward and inverted indexes that record which words appear on which pages, and where. | Still the entry ticket: Google says a page must be indexed and eligible to show a snippet in Search to appear as a supporting link in AI Overviews or AI Mode. Pages blocked by robots.txt or noindex stay out. | Same |
| RecallStep 6 | Look up each query word in the inverted index and keep candidate pages with OR / AND. | One question may be expanded into several related sub-queries (Google calls this query fan-out), so a page can be retrieved for sub-questions the user never typed. Semantic matching widens the candidate pool further. | Changed |
| ScoringStep 7 | BM25 scores the whole page using field weights, word rarity and page length. | Candidates still get ordered, but the unit is often a passage rather than the whole page, and the model then combines several passages into one answer. The exact weighting is not public. | Changed |
| Ranking & resultStep 8 | Pages are sorted into a list of ten blue links; the title and snippet earn the click. | The result is a generated answer, and pages appear as cited links or mentions rather than as a numbered position. Which sources are cited, and how many, can vary from answer to answer. | Changed |
Four things to do in your on-page writing
- Make each passage stand on its own.Passages are often retrieved and cited one at a time. Name the subject (the product, brand or concept) instead of relying on “it” or “this” from an earlier paragraph.
- Lead with the answer under a descriptive heading.Use a heading that says what the section answers, and put the direct answer in the first sentence before the detail.
- State entities, numbers, dates and units.Say which product, how much, when and in what unit, instead of “fast”, “recently” or “a lot”. Specific statements are easier to quote accurately.
- Keep the basics from Steps 1–6.Core content in the served HTML, not blocked by robots.txt or noindex, indexable, with a clear Title and H1. A page that is never retrieved cannot be cited.
Note: this simulator models lexical matching (BM25) only. It does not model embeddings, query fan-out or answer generation.
Search glossary & parameter guide
Plain-English definitions linking each metric, index table, and slider in this simulator to everyday SEO work.
| Term / parameter | Plain-English meaning | What it does here | What to do on your site |
|---|---|---|---|
| robots.txt / noindex | Rules controlling crawler access vs whether a page is indexed. | Decides in Step 1 whether a page moves on to Steps 2 through 8. | Keep revenue and guide pages open to both; block admin pages and noisy filters. |
| Field weights | Multipliers giving Title and headings higher relevance weight than body copy. | Defaults in Step 7 to Title ×5.0, H1 ×3.5, H2–H6 ×2.0, Body ×1.0 with live sliders. | Get Title and H1 right first, then structure supporting sections with H2–H6. |
| Stop words & folding | Dropping filler words and normalizing case plus common plurals/inflections. | Keeps the index lean in Step 3 so queries match indexed page words cleanly. | Write naturally for human readers instead of forcing stiff exact-match phrasing. |
| Forward index & length | Per-page list of words, mention counts, and total indexed word count (called dl). | Flags body stuffing in Step 4 and feeds page length into Step 7 scoring. | Check for skewed repetition and trim unrelated tangents that bloat page length. |
| Inverted index | Word → Matching Pages dictionary that also counts how many pages include each word (called df). | Drives Step 5’s connection graph and Step 6’s fast candidate lookup. | Make sure your pages actually use the phrasing buyers type into search. |
| Word rarity (IDF) | Inverse Document Frequency: words found on fewer pages are rarer and carry more weight. | Calculated in Step 7 so specific words outscore generic filler. | Use specific product attributes and long-tail terms where competition is thinner. |
| BM25 | A classic relevance formula used by search engines (BM stands for "Best Matching"). A page scores higher when query words appear more often, are rarer across pages, and sit in a focused page, but repeating one word only helps up to a ceiling. | Step 7 computes BM25 for Title, H1, H2–H6, and Body separately, multiplies each by its field weight, and adds them up; Step 8 sorts by the total. | Put the query naturally in Title and H1 and keep body copy focused instead of repeating keywords. |
| TF-IDF | Term frequency (TF) times inverse document frequency (IDF): a word scores higher the more often it appears on this page and the rarer it is across all pages. It is an older, simpler classic scoring method than BM25. | Not used directly here. Step 7’s BM25 can be read as an improved TF-IDF: it also uses word frequency and rarity, but caps the gain from repeating a word (k1) and adjusts for page length (b). | When you see advice about “TF-IDF optimization”, read it as “state the core terms clearly, in the places that matter”, not as hitting a fixed keyword density. |
| BM25 k1 & b | k1 caps repetition gain (ceiling k1 + 1); b controls how much long pages are penalized. | Defaults in Step 7 to k1 = 1.2, b = 0.75, showing why repeating one word ten times stalls. | Cover subtopics thoroughly instead of repeating the same keyword in body copy. |
| OR / AND recall | OR matches any query word; AND requires a page to include every query word. | Sets candidate strictness in Step 6; matching every word earns a +25% bonus in Step 7. | For multi-word searches like "shopify speed", address every part of the query clearly. |
Frequently asked questions
Straight answers on how this simulator compares to live Google Search, why stuffing stalls, snippet rewrites, indexing checks, and whether any of this still applies to AI search.
How close is this simulator to real Google Search?
Everything here runs a genuine 8-step client-side search flow (crawl gating, boilerplate stripping, tokenization, forward and inverted indexes, recall, multi-field BM25 scoring, and snippet selection) rather than canned animations. Real Google Search adds hundreds of web-scale signals on top, including backlinks (PageRank), neural semantics (BERT and MUM), user engagement (Navboost), Core Web Vitals, and E-E-A-T (experience, expertise, authoritativeness, and trust). Use this bench to see how on-page fundamentals actually work.
Why does Stuffed Blog lose to SEO Guide even though it repeats the keyword more often?
Three things work against Stuffed Blog at once. First, field weights: SEO Guide places query terms cleanly in Title (×5.0) and H1 (×3.5), while Stuffed Blog gets zero from both. Second, repetition caps out (k1 = 1.2): each extra mention adds less, and the gain can only approach the k1 + 1 = 2.2 ceiling (with default settings and an average-length page, 10 mentions give about 1.96), which cannot make up for missing headings. Third, length penalty (b = 0.75): Stuffed Blog’s unrelated rambling pushes its word count above average, shrinking its body score further.
For a Shopify or WordPress site, which HTML elements should I fix first?
Work in order of ranking impact: first verify crawlability and indexation (robots.txt open, no stray noindex, clean canonical); second sharpen Page Title and your single H1 around the main search term; third organize H2 and H3 subheadings around buyer questions and specs; fourth write a concise, query-matched Meta Description to improve click-through rate.
Why does Google sometimes ignore my Meta Description and pull body text instead?
Google picks whichever snippet best shows the searcher that your page answers their specific query. If your Meta Description is vague or leaves out the searched terms, Google skips it and grabs the best-matching excerpt from the body instead. Try switching preset queries in Step 8 (Ranking) above to see the snippet source flip live.
My page has been live for weeks but never shows up in Google. Where should I check?
Check in step order. Start at Step 1 (Crawl): run Google Search Console URL Inspection to check robots.txt, meta noindex, HTTP status, and canonical selection. Next check Steps 2 and 6 (Extract and Recall): make sure key copy sits in raw HTML and actually uses the words people search. Once indexation and recall are solid, work on Step 7 relevance (Title, H1, and content depth) against page-one competitors.
Can I paste my own HTML or copy to test? Does anything get uploaded?
Yes. Click "Edit raw HTML" in Step 1 (Crawl) or paste your Title, H1, description, and body directly into Step 8’s optimization bench. Every parser, tokenizer, and scoring step runs locally inside your browser. Nothing leaves your device.
Does this still apply to AI search?
The retrieval half does. Google’s AI Overviews and AI Mode, and AI assistants that search the web, usually retrieve pages and passages before generating an answer, and that retrieval still needs a page that can be crawled, indexed and matched, as in Steps 1–6. What changes comes after: semantic matching alongside keywords, one question expanded into several sub-queries, passages used instead of whole pages, and being cited in an answer instead of holding a numbered position. This simulator models keyword matching (BM25) only, not those AI steps.
About this tool & Jindouyun SEO x GEO
Built and maintained by the Jindouyun SEO x GEO team so ecommerce founders, Shopify merchants, and content teams can see how on-page signals actually affect retrieval and ranking. Reach out if you spot an edge case to add, or want a hands-on technical SEO and AI search (GEO) review for your site.