On-Page SEO: What Google Reads and Ignores

On September 14, 2026, we typed “seo on page” into Google. The search ran on Google in Brazil, where we are based. At the top, an AI answer handed over a list. Title, description, headings, helpful content, internal links, images. ¹

Scroll down the results, and the list repeats itself. The order changes, and so do the numbers. That same day, we read the thirteen addresses that shared the first page of that search. They came from the live results and from Semrush, and most of them are in Portuguese. The recommended title length ran from 40 to 70 characters. The description ran from 105 to 160. ²

Google writes something else. There is no length limit for the title or for the description. The result gets truncated when needed, to fit the width of the screen. ³ ⁴

One question was missing from the list. Which of these items does Google write that it reads? Which does it write that it ignores? And which rule circulates with no document at all? This piece goes item by item, each one with the official page behind it.

What on-page SEO is

Think of a box of cold medicine on a pharmacy shelf. Before anyone opens it, the box has already told you the name, what it treats, and what is inside. If the box promises one thing and holds another, it has lost the trust of whoever read it.

A page does the same job. It tells Google and AI answers what it is and what topic it covers. It says so through the title, the description, the headings, and the text. It also says so through links, images, the address, and the labels hidden in the code. This is called on-page SEO.

Two neighbors complete the picture. What other sites say about your page belongs to off-page SEO. The crawler’s ability to reach the page and read what is on it belongs to technical SEO. This piece covers only what the page declares about itself.

Three questions for each item

For each item below, I ask three questions of what Google has published. Does it write that it uses the item? Does it write that it ignores it? Or does the rule circulate with no source?

The first two answers have an address in the documentation. The third is the one that shows up most in the lists, and it gets its own section near the end.

The title and the title link

The blue, clickable text on each result has a name in Google’s documentation. This is called a title link. Google builds that link on its own, from nine sources. ³

The first source is the title element, the <title> in the code. Next come the visible title on the page, headings such as the H1, and the og:title tag. Then the large, prominent text, the rest of the text, and the anchor text on the page itself. The list closes with the text of outside links pointing to the page, and the site’s structured data. ³

In practice, the title element almost always wins. In September 2021, Google wrote that it was used about 87% of the time. ⁵ In our search on September 14, all nine titles on the first page came from the title element. Two were cut off with an ellipsis, and one lost the site name. Google translated one more from English into Portuguese. ⁶

The documentation lists the cases where Google swaps the text. Half-empty titles, obsolete titles, and titles that do not describe the page. Titles repeated across a group of pages, pages with no clear main title, and titles in another language. ³

Our own site fell into that last case. Our Portuguese homepage served a title in English, and the story is in the piece on SEO for WordPress. ⁷ Google writes that when the title’s language does not match the text, it may choose a different one. ³

What Google asks for fits on one line. One title per page, descriptive, concise, distinct from the others, and in the language of the text. ³

The meta description and the snippet

Under the blue title, two or three lines of summary appear. This is called a snippet. The text you suggest for that summary lives in the code, and it is called a meta description.

Google writes that it builds the snippet mainly from the page content. The meta description comes in sometimes, when it describes the page better than the text itself. ⁴

We checked this on the same search. Of the nine summaries on the first page, eight did not come from the page’s meta description. In all eight, the text shown came from the article itself. The ninth was the meta description, translated from English. ⁶

The documentation explains the swap. The snippet is chosen for the specific search, and it can change from one search to the next. ⁴ And Google wrote in 2009 that it does not use the meta description to rank the page. ⁸

It is still worth the effort. Google’s help asks for a unique description on every page, starting with the homepage and the most popular pages. ⁴ The rest of the work sits in the first sentences of each section, because the summary comes mainly from the text.

Headings

On headings, Google writes two things that seem to clash.

The first is in the starter guide. Heading order is great for screen readers, and it makes no difference to Search. There is also no magic number of headings per page. ⁹

The second is on the title link page. Google asks for the main title to stand out from the rest, for example in the first visible H1 on the page. ³ When several large headings carry the same weight, it may take the first one. ³

Both statements fit together. The guide talks about how the tags are numbered and how many there are. The other page talks about making clear which text is the main title.

Five of the thirteen articles on the first page recommend a single H1 per page. ² In the Google pages I read, that rule does not appear.

Headings have a third job, and it is the least visible one. A Google patent, granted in 2022 and still active, describes the path of headings down to a passage. It is the sequence from the page title to the heading right above the answer. That path feeds the score a passage gets as the answer to a question. ¹⁰ A patent does not prove implementation. The practical reading, with the prompt for writing headings this way, is in titles as addresses.

The text

Google sums up the job of the text in one recommendation. Use the words people would use to look for your content. Put those words in prominent places: the title, the main heading, image alt text, and link text. ¹¹

Three limits come from the documentation itself. The length of the text alone does not matter for ranking, and there is no minimum word count. ⁹ Repeating the same words until they sound unnatural is keyword stuffing, and that is in the spam policies. ¹² And Google’s language systems understand a page even without the exact words of the search. ⁹

What remains is saying what the page is about. Ask three people to summarize your page in one sentence. The name that survives all three summaries is the topic of the page, for whoever read it. This is called salience.

The idea has a published origin. In 2014, a researcher from Carnegie Mellon and one from Google Research defined the task. Each entity in a document gets a relevance score. ¹³ Google Cloud’s Natural Language API delivers that score today, from 0 to 1, for each entity. The documentation says it measures how central the entity is to the whole text. ¹⁴

The caveat matters. That API is a cloud product, and nothing in its documentation says Search uses the same number. It works as a mirror. If the most central name in your text is the competitor you cited as an example, the page is talking about them.

Internal links

The clickable text of a link tells the reader and Google something about the destination page. ¹⁵ The documentation asks that every page you care about get a link from at least one other page on the site. And it repeats the line from the headings section: there is no magic number of links. ¹⁵

Which page each link should point to is a topic with a piece of its own, on internal links.

Images and alt text

Alt text describes the image for people who cannot see it. Google writes that it uses this text together with computer vision and the page content. That is how it works out what the image is about. ¹⁶

The file name gives very light clues, in the documentation’s own words. ¹⁶ Stuffing alt text with keywords appears on the same page as a risk of the site looking like spam. ¹⁶ And the starter guide asks for the image to sit near the text that discusses it. ⁹

The URL

Google recommends readable words in the URL, in your audience’s language, separated by hyphens. ¹⁷ And it writes that a keyword in the domain or the path, by itself, has hardly any effect beyond breadcrumbs. ⁹

The URL works for the person. They understand the link before they click.

Structured data

Think of the spec label stuck on an appliance box. Brand, model, and voltage sit in fixed fields, with no need to open the manual. On a page, that label is a block of code with fields the machine reads directly. This is called structured data.

Google writes that it uses this markup to understand the page content. ¹⁸ The rule that matters most is in the guidelines. The markup has to represent what the reader sees, and the guideline says not to mark up invisible content. ¹⁹ Even correct markup does not guarantee a rich result. ¹⁹

In my book SEO de Entidade, published in Portuguese, I called this markup a declaration of identity. The chapter’s verdict: do it for the right reason, and never as a switch for citation. ²⁰

The canonical URL

Sometimes the same content answers at two addresses. Google groups the copies and picks one to show. The address it picks has a name. This is called the canonical URL. ²¹

You can state your preference, and the documentation gives the weight of each method. A redirect is a strong signal. The rel="canonical" tag is, too. A sitemap is a weak signal. ²² The documentation also asks the canonical page to point to itself. ²²

Google calls that preference a hint, and warns that it may choose another page. ²¹ Duplicate content inside your own site does not trigger a manual action. The guide calls it inefficient. ⁹

What Google writes that it ignores

There is an official list, and it is short.

The keywords tag, the meta keywords, has no effect at all on indexing or ranking. ²³ Google stopped reading it because it was abused, and put that on record in 2009. ⁸ None of the thirteen articles on the first page mentions this tag. ²

Google also leaves out the HTML lang attribute. It detects the language from the text on the page. ²³

The starter guide also has a section of things not worth focusing on. The number and order of headings are on it, and so is content length. ⁹ And an acronym that shows up in three of the thirteen articles, E-E-A-T. The guide asks whether it is a ranking factor, and answers no. ² ⁹

The rules that circulate with no document

The rules below appeared on the first page for “seo on page” on September 14. Next to each one sits what Google’s documentation says about it. ²

What circulates	What Google writes
A title of 40 to 70 characters	There is no limit on the title element; the title link is truncated to fit the screen ³
A description of 105 to 160 characters	There is no limit on the meta description; the snippet is truncated to fit the screen ⁴
A single H1 per page	Heading order does not matter to Search, and there is no ideal number of headings ⁹
A text of at least 500 words	Length alone does not matter for ranking ⁹
Three uses of the keyword every 500 words	No published number; excessive repetition is spam ⁹ ¹²
A keyword required in the URL	By itself, hardly any effect beyond breadcrumbs ⁹

One of them comes attributed to Google itself. One of the articles says Google recommends at least 500 words. ² Google’s guide says there is no minimum. ⁹

None of these rules was born in bad faith. The character limit, for example, tries to predict where the screen will cut. The problem shows up when a rule of convenience becomes a ranking factor in the text of whoever teaches it.

What the page declares to AI

The AI answers in Google Search, AI Overviews and AI Mode, work on top of the search index. ²⁴ Google writes that there are no additional requirements to appear in them. ²⁵

The technical condition fits on one line. The page has to be indexed and eligible to appear in Search with a snippet. ²⁵ This is where on-page SEO touches AI by a concrete thread. Tags that limit the snippet, like nosnippet and max-snippet, are the control a page has over what appears in these features. ²⁵

On that same page, the list of best practices reads like the outline of this piece. Important content in text form, internal links, quality images, and structured data that matches the visible text. ²⁵

Google’s guide to generative AI, updated in July 2026, rules out three shortcuts. Special files for AI, text chopped into tiny pieces, and special schema markup. ²⁴ For Google’s AI, the page declares the same things, in the same places.

I am talking here about Google’s AI answers, which publish their rules. Other assistants have their own documentation, outside this piece.

The twenty-minute check

Pick the page on your site that sells the most. Open it, press Ctrl+U to see the source code, and use the browser’s find tool.

Search for <title>. Read it out loud. Does the text describe this page, and is it in the page’s language? ³
Search for description. Does the description exist, and does it talk about this specific page? ⁴
Search for <h1>. Is the main title the text that stands out most on the screen? ³
Search for alt=. Does every important image have a description of what it shows? ¹⁶
Search for canonical. Is the address it points to the page’s own address? ²²
Search for ld+json. Does everything the markup declares appear on the visible page? ¹⁹
Search for keywords. If the tag exists, it can stay. It neither helps nor hurts. ²³
Look the page up on Google. Compare the title and summary shown with what you wrote. If they changed, the documentation’s list of cases tells you why. ³ ⁴

The check covers one page. The full order, from your measurement tools to the business, is in the SEO audit.

From now on, when a list of on-page factors lands on your desk, you will ask which Google page each line came from. The lines with no address go to the bottom of the pile.

Open the page that sells the most today, and press Ctrl+U.

Frequently asked questions

What is the right length for the title and the meta description?

Google sets no limit for either one. The title shown and the summary are cut when needed, to fit the width of the screen. ³ ⁴ The documentation asks for a title that is descriptive and concise, with no needlessly long text. ³

Does Google still use the meta keywords tag?

No. The documentation says this tag has no effect at all on indexing or ranking. ²³ If your site has it, it can stay.

Can I have more than one H1 on a page?

Google’s guide says the order and number of headings make no difference to Search. ⁹ The title link page asks for the main title to stand out, for example in the first visible H1. ³ Make it clear which one is the main title. The rest serve the reader.

Do I need to do anything different to show up in Google’s AI answers?

Google writes that AI Overviews and AI Mode have no additional requirements or special optimizations. The page has to be indexed and eligible to appear with a snippet. ²⁵ Special AI files and special schema markup are absent from the list of requirements. ²⁴

Notes and sources

Live reading of the Google Brazil results page for “seo on page,” September 14, 2026, at https://www.google.com.br/search?q=seo+on+page&hl=pt-BR&gl=br&pws=0, in the author’s browser. The search ran on Google in Brazil, where we are based, with the interface in Portuguese. The AI Overview listed, in Portuguese, as its main elements the title tag, meta description, heading tags, helpful content, internal links, and image optimization, with HubSpot and Semrush among its sources. Accessed September 14, 2026.
SEO Genome survey, September 14, 2026: the top ten organic results in Semrush, Brazil database, for “seo on page,” added to the nine on the first page of Google Brazil in note 1, for a total of thirteen addresses. All texts were read the same day; most are in Portuguese. Recommended title length: 40 to 60 characters at https://agenciamutum.com/blog/seo-on-page-off-page-e-tecnico/ and 70 at https://www.ekyte.com/guide/pt-br/blog/seo-on-page-e-off-page-qual-diferenca-e-como-funciona/, with the others in between. Recommended description length: about 105 characters at https://www.semrush.com/blog/on-page-seo/ and 160 at https://br.hubspot.com/blog/marketing/seo-on-page and https://www.magazord.com.br/conteudo/blog/seo-on-page, with the others in between. Recommending a single H1 per page: https://www.conversion.com.br/blog/seo-on-page/, https://mailchimp.com/pt-br/resources/all-about-on-page-seo/, the eKyte text, https://www.seohappyhour.com/blog/fazer-seo-on-page/, and the Magazord text. The HubSpot text suggests three occurrences of the keyword every 500 words; the Magazord and Conversion texts treat the keyword in the URL as necessary for ranking; the text at https://ebaconline.com.br/blog/seo-on-page-e-off-page-entenda-as-diferencas attributes to Google a recommendation of texts with at least 500 words. E-E-A-T appears in the Conversion and Agência Mutum texts and at https://www.organicadigital.com/blog/seo-on-page-passos-para-otimizar/. None of the thirteen mentions the keywords meta tag. The other addresses read were https://pt.semrush.com/blog/seo-on-page/, https://br.hubspot.com/blog/marketing/seo-on-page-off-page, and https://fundamentosdeux.com/post/se0-on-page-como-otimizar-sua-pagina-para-rankear-bem-nos-mecanismos-de-busca-616.htm. Accessed September 14, 2026.
Google Search Central, “Influencing your title links in search results,” https://developers.google.com/search/docs/appearance/title-link, page last updated December 10, 2025. Accessed September 14, 2026. Passages: “Google uses a number of different sources to automatically determine the title link”; “While there’s no limit on how long a <title> element can be, the title link is truncated in Google Search results as needed, typically to fit the device width”; the list of nine sources, from “Content in <title> elements” to “WebSite structured data”; the suggestion to make the main title stand out, “putting the title text in the first visible <h1> element on the page”; the case of several large headings, where Google “may use the first heading as the text for the title link”; the list of common issues (half-empty, obsolete, inaccurate, repeated across a group of pages, no clear main title, in another language); and “If Google determines that a <title> element does not match the writing system or language of the page’s primary content, we may choose a different text as the title link.”
Google Search Central, “Control your snippets in search results,” https://developers.google.com/search/docs/appearance/snippet, page last updated April 20, 2026. Accessed September 14, 2026. Passages: “Snippets are primarily created from the page content itself. However, Google sometimes uses the meta description HTML element if it might give users a more accurate description of the page”; “Google Search might show different snippets for different searches”; “There’s no limit on how long a meta description can be, but the snippet is truncated in Google Search results as needed, typically to fit the device width”; and “at the very least, create a description for the critical URLs like your home page and popular pages”.
Google Search Central Blog, “More information on how Google generates titles for web page results,” September 17, 2021, https://developers.google.com/search/blog/2021/09/more-info-about-titles. Accessed September 14, 2026. Passage: “title elements are now used around 87% of the time, rather than around 80% before.”
SEO Genome survey, September 14, 2026, of the nine organic results in note 1. The title and summary shown for each result were compared with the <title> and the meta description in the HTML downloaded the same day. Titles: five identical to the <title> (HubSpot, Semrush in Portuguese, EBAC, Orgânica Digital, eKyte), two cut off with an ellipsis (Conversion and Magazord), one without ” | Mailchimp,” and one translated from English by Google (Semrush in English). Summaries: in eight, the text shown was not the meta description and was found in the body of the article (seven in the downloaded HTML; the Magazord one, which loads its text by script, in the browser); in the ninth, Semrush in English, the summary was the translation of the meta description, a sentence absent from the body. Accessed September 14, 2026.
SEO Genome internal record, “Diagnóstico do site seogenome.com” (seogenome.com site diagnosis), August 31, 2026, section 2.6: the <title> served on the Portuguese homepage was in English. The site then served Brazil, where we are based, in Portuguese. The full case is in this site’s piece on SEO for WordPress.
Matt Cutts, “Google does not use the keywords meta tag in web ranking,” Google Search Central Blog, September 21, 2009, https://developers.google.com/search/blog/2009/09/google-does-not-use-keywords-meta-tag. Accessed September 14, 2026. Passages: “Because the keywords meta tag was so often abused, many years ago Google began disregarding the keywords meta tag” and “Even though we sometimes use the description meta tag for the snippets we show, we still don’t use the description meta tag in our ranking.” The page now warns that some of its information may be outdated, and restates that “Google still doesn’t support the keywords meta tag.”
Google Search Central, “SEO Starter Guide,” https://developers.google.com/search/docs/fundamentals/seo-starter-guide, page last updated December 10, 2025. Accessed September 14, 2026. Passages: “Having your headings in semantic order is fantastic for screen readers, but from Google Search perspective, it doesn’t matter if you’re using them out of order”; “There’s also no magical, ideal amount of headings a given page should have”; “The length of the content alone doesn’t matter for ranking purposes (there’s no magical word count target, minimum or maximum”; “the keywords in the name of the domain (or URL path) alone have hardly any effect beyond appearing in breadcrumbs”; “Google’s language matching systems are sophisticated and can understand how your page relates to many queries, even if you don’t explicitly use the exact terms in them”; images “near text that’s relevant to the image”; on content at several addresses, “It’s inefficient, but it’s not something that will cause a manual action”; “Excessively repeating the same words over and over (even in variations) is tiring for users, and keyword stuffing is against Google’s spam policies”; “Google Search doesn’t use the keywords meta tag”; and, on E-E-A-T as a ranking factor, “No, it’s not.”
Patent US 11,409,748 B1, “Context scoring adjustments for answer passages,” Google LLC, inventors Nitin Gupta, Srinivasan Venkatachary, Lingkun Chu, and Steven D. Baker, https://patents.google.com/patent/US11409748B1/en. Priority January 31, 2014, filed March 16, 2018, granted August 9, 2022; status read on the page: active, expiring July 31, 2035. Accessed September 14, 2026. The text describes “a heading vector that describes a path in the heading hierarchy from the root heading to the respective heading” and a context signal that “may be, in part, query-independent”, used to score candidate answer passages for question-form queries. The patent does not say the mechanism runs today, or that a heading level improves a page’s position.
Google Search Central, “Google Search Essentials,” https://developers.google.com/search/docs/essentials, page last updated December 10, 2025. Accessed September 14, 2026. Passage: “Use words that people would use to look for your content, and place those words in prominent locations on the page, such as the title and main heading of a page, and other descriptive locations such as alt text and link text.”
Google Search Central, “Spam policies for Google web search,” https://developers.google.com/search/docs/essentials/spam-policies, page last updated August 28, 2026. Accessed September 14, 2026. Passage: “Keyword stuffing refers to the practice of filling a web page with keywords or numbers in an attempt to manipulate rankings in Google Search results”, with the example “Repeating the same words or phrases so often that it sounds unnatural.”
Jesse Dunietz (Carnegie Mellon University) and Dan Gillick (Google Research), “A New Entity Salience Task with Millions of Training Examples,” EACL 2014, volume 2, pages 205 to 209, https://aclanthology.org/E14-4040/. Accessed September 14, 2026. From the abstract, read in the PDF: “we introduce the task of entity salience: assigning a relevance score to each entity in a document”, with the labeled corpus generated from documents and the summaries that accompany them. The three-summary exercise is the author’s reading of that design.
Google Cloud, Natural Language API, REST reference, “Entity” object, https://docs.cloud.google.com/natural-language/docs/reference/rest/v1/Entity, page last updated June 13, 2025. Accessed September 14, 2026. Passage: “The salience score for an entity provides information about the importance or centrality of that entity to the entire document text”, on a scale of 0 to 1.0. The documentation is for the cloud product and does not deal with Search.
Google Search Central, “Link best practices for Google,” https://developers.google.com/search/docs/crawling-indexing/links-crawlable, page last updated December 10, 2025. Accessed September 14, 2026. Passages: anchor text “tells people and Google something about the page you’re linking to”; “Every page you care about should have a link from at least one other page on your site”; and “There’s no magical ideal number of links a given page should contain.”
Google Search Central, “Google image SEO best practices,” https://developers.google.com/search/docs/appearance/google-images, page last updated March 2, 2026. Accessed September 14, 2026. Passages: “Google uses alt text along with computer vision algorithms and the contents of the page to understand the subject matter of the image”; “the filename can give Google very light clues about the subject matter of the image”; and, on filling the attribute with keywords, that it “may cause your site to be seen as spam”.
Google Search Central, “URL structure best practices for Google Search,” https://developers.google.com/search/docs/crawling-indexing/url-structure, page last updated December 10, 2025. Accessed September 14, 2026. Passages: “use readable words rather than long ID numbers in your URLs”, “Use words in your audience’s language in the URL”, and the recommendation of hyphens to separate words.
Google Search Central, “Introduction to structured data markup in Google Search,” https://developers.google.com/search/docs/appearance/structured-data/intro-structured-data, page last updated December 10, 2025. Accessed September 14, 2026. Passage: “Google uses structured data that it finds on the web to understand the content of the page”.
Google Search Central, “General structured data guidelines,” https://developers.google.com/search/docs/appearance/structured-data/sd-policies, page last updated July 10, 2026. Accessed September 14, 2026. Passages: “Don’t mark up content that is not visible to readers of the page”; “Your structured data must be a true representation of the page content”; and “Google does not guarantee that your structured data will show up in search results, even if your page is marked up correctly”.
Daniel Sócrates, SEO de Entidade (published in Portuguese), 2026, chapter 11, “Os cinco mitos contra a evidência” (The five myths against the evidence), charge 4. The chapter treats structured data markup as a declaration of identity, and the verdict is to do it for the right reason, and never as a switch for citation.
Google Search Central, “What is canonicalization,” https://developers.google.com/search/docs/crawling-indexing/canonicalization, page last updated August 20, 2026. Accessed September 14, 2026. The page describes how duplicate pages are grouped and the canonical is chosen, and says: “indicating a canonical preference is a hint, not a rule.”
Google Search Central, “How to specify a canonical URL with rel=“canonical” and other methods,” https://developers.google.com/search/docs/crawling-indexing/consolidate-duplicate-urls, page last updated July 10, 2026. Accessed September 14, 2026. The methods, “in order of how strongly they can influence canonicalization”: redirects, “A strong signal”; rel="canonical", “A strong signal”; sitemap, “A weak signal”. And the best practice “Do include a rel=“canonical” link on the canonical page itself (also known as a self-referential canonical).”
Google Search Central, “Meta tags and attributes that Google supports,” https://developers.google.com/search/docs/crawling-indexing/special-tags, page last updated December 10, 2025. Accessed September 14, 2026. In the section on tags Google does not support: “The meta-keyword tag is not used by Google Search, and it has no effect on indexing and ranking at all”; and, on the lang attribute, “Google Search detects the language of a page based on the textual content of the page.”
Google Search Central, “Optimizing your website for generative AI features on Google Search,” https://developers.google.com/search/docs/fundamentals/ai-optimization-guide, page last updated July 10, 2026. Accessed September 14, 2026. Passages: generative AI features in Search “are rooted in our core Search ranking and quality systems” and highlight content “from our Search index”. In the section on what you do not need to do: “You don’t need to create new machine readable files, AI text files, markup, or Markdown”; “There’s no requirement to break your content into tiny pieces for AI to better understand it”; and “Structured data isn’t required for generative AI search, and there’s no special schema.org markup you need to add.”
Google Search Central, “AI features and your website,” https://developers.google.com/search/docs/appearance/ai-features, page last updated December 10, 2025. Accessed September 14, 2026. Passages: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary”; “a page must be indexed and eligible to be shown in Google Search with a snippet”; the list of best practices, with internal links, “important content is available in textual form”, quality images, and structured data that “matches the visible text on the page”; and, to limit what appears, “use nosnippet, data-nosnippet, max-snippet, or noindex controls.”

On-Page SEO: What Your Page Tells Google, Item by Item