SEO — what actually decides whether Google shows your site

SEO is bought differently from everything else a company buys. With a website you see the result and can judge it. With SEO you buy a promise about the behaviour of a system none of us can see inside — and an invoice that runs whether or not the promise holds.

This article is about the parts you can check yourself: what Google says about its own search engine in its own documentation, and what shows up in the data when somebody actually looks. Every figure about our own site below is ours, measured on our Polish domain, and none of it is flattering.

Start with the sentence Google wrote about this industry

On the page "Do you need an SEO?" in Google's own Search Central documentation there is a sentence that settles most sales conversations about search:

> No one can guarantee a #1 ranking on Google.

Not "it is difficult to guarantee", not "it depends on the sector" — no one. The same page lists the warning signs of a bad supplier, again in Google's words: a firm that contacts you out of the blue by email, a firm promising guaranteed rankings, a firm claiming a "special relationship with Google", a firm that is secretive or will not explain clearly what it intends to do, and a firm offering link popularity schemes or submission to thousands of search engines.

The document is public, maintained and free. If you take one thing from this article, let it be that the maker of the system has described how to recognise somebody who claims to beat it — and that the description needs no login.

The rest of this text covers what is left once the promises are set aside: the four states your pages can be in, the difference between being indexed and being visible, and which technical factors Google calls a signal rather than folklore.

Before you worry about position, check that Google has your pages at all

Conversations about search almost always begin with position. That is one step too far. The question before it is whether Google holds your pages in the first place — and the answer can be a surprise.

We checked every blog post in our sitemap in Search Console. The result on 9 September 2026, as shares:

• 20% indexed
• 51% "crawled — currently not indexed" — Google fetched the page and did not index it
• 16% "discovered — currently not indexed" — Google knows the address but has not fetched it
• 12% unknown to Google at all
• 1% excluded by a noindex tag, as intended

One page in five. The site has been publishing steadily for years.

The two middle states are the most widely misunderstood, so it is worth reading what Search Console help says about them. On the first:

> The page was crawled by Google but not indexed. It may or may not be indexed in the future; no need to resubmit this URL for crawling.

No need to resubmit. That matters, because the advice a business owner usually receives in this situation is "open Search Console and click Request Indexing". The maker of the tool says plainly that this achieves nothing. The page was looked at and declined — that is a judgement, not a fault, and clicking does not change a judgement.

About the second state Google says something different and equally concrete:

> The page was found by Google, but not crawled yet. Typically, Google wanted to crawl the URL but this was expected to overload the site; therefore Google rescheduled the crawl.

Here the problem is not the content but the risk of overloading your server. It is the only one of the four states where hosting and performance are a real cause rather than an excuse.

The practical conclusion for somebody paying for SEO: these four states are the first thing to look at in a report. If half the site sits in "crawled — currently not indexed", as ours did, no amount of work on positions has anything to lift, because those pages are not in the set Google draws results from.

What to do about pages Google looked at and declined

If resubmitting achieves nothing, the question is what does. We answer it with what we did about our own rejected half, rather than with general advice.

The most common cause in our case was several texts answering the same question. Three articles about search — the one you are reading was merged from exactly such a trio. Together they ran to over eleven thousand words and shared five chapters: keyword research, content optimisation, speed, local visibility and measurement. To a search engine they were three variants of one page, each weakening the other two.

What we do about it, in order:

1. Find the repetitions. Not by title, by heading — if the same chapter appears in three texts, you have three texts about one thing.
2. Merge into one and redirect the rest permanently. An address that disappears without a redirect takes with it every link anyone ever pointed at it.
3. Check that the text answers a question people actually ask, rather than one that seemed interesting internally. That is the next section and it is the most common cause of all.
4. Give the page internal links. An address nothing points to may simply never be discovered — twelve per cent of ours are in that state.

What is not on that list: asking for reindexing, adding keywords to an existing text, and publishing more posts in the hope that one catches on. The first is ineffective by documentation, the second makes the text worse, and the third is the subject of the next section.

Being indexed is not the same as being visible

Indexation and visibility are two different measurements, and confusing them is the most common error in SEO reporting. A page can be indexed and go a whole quarter without being shown to anyone.

Our own figures for 90 days:

• blog posts: 14% of addresses drew any impression in Google
• practical tools — calculators, reports, downloadable templates — 54%

Same domain. Same team, same CMS, same technique, same hosting. The hit rate differs almost fourfold.

It is worth being clear about what this pair of numbers does not prove. It is one site and one sample; there are far fewer tools than posts, so a single successful calculator weighs more in that share than a single successful article. The direction is too pronounced to dismiss, though: something that calculates or shows you an answer matches a query more often than a text describing a topic.

And from that follows the part the industry finds least comfortable. The most commonly sold prescription is "we must publish regularly for Google to notice you". Our own data says regularity is not the mechanism — we publish regularly and have done for years, and 86% of posts were not shown in results even once in a quarter. It does not mean content fails. It means content works when it answers a question somebody genuinely puts to a search engine, and not when it appears in a publishing calendar.

A phrase nobody types is worse than no phrase at all

The most expensive mistake in search is not technical. It is writing for a word your customers do not use.

We made it ourselves. An article about tools for building your own site targeted a compound noun that sounds professional in our market and that, checked in Google's Keyword Planner, has no measured search volume at all — not rare, simply not how people ask. The everyday synonym our market actually types carries about 2,550 searches a month.

The text was sound by every measure used to judge SEO work, and it was invisible, because it answered a question asked in somebody else's words. We rewrote it and changed the address. We mention it because this is precisely the error that is hard to see from inside: it sounds professional, so nobody questions it.

A second mistake from the same family is easier to make and harder to notice: you can compete with yourself. It happened to us in a single day. In the morning we rewrote a text about goals and metrics and gave it a secondary phrase; in the afternoon we rewrote a text about analytics tools for which that exact phrase was the main one. Neither decision was wrong on its own — the error lived between them, because each was checked against its own department rather than against the whole site.

Hence the rule we have followed since: check a phrase against the entire site before assigning it, not against one section. Two of your pages aiming at the same query do not add up — they split the signal, and Google picks one, usually not the one you would have picked.

How to check this yourself, without paid tools:

• Google's Keyword Planner is free and shows approximate volumes. Set the country and language before you read anything.
• Check synonyms, not only your phrase. People describe the same thing in several ways and it is not always the industry way.
• Separate a question about knowledge from a question about a supplier. "SEO" is asked by somebody who wants to understand. "SEO agency" and "SEO pricing" are asked by somebody who wants to buy — and that second kind belongs on your service pages, not your blog. Putting a blog article against your own sales page is competing with yourself.
• Long phrases are often worth more than short ones. Fewer queries, but a clear intent and less competition.

Titles and descriptions — what they do and what they do not

The page title and meta description are the text Google shows in results. They are worth writing carefully, and worth understanding.

They do not raise your position. They decide whether somebody clicks your result once they see it. That is work on traffic, not on ranking, and the distinction matters when you read a report.

Practically: neither has a character limit — Google's documentation states plainly that there is no maximum length for a title or a description, and that the result is truncated "as needed, typically to fit the device width". The familiar 60-and-160 rule is industry folklore, not something Google wrote. The useful conclusion is not a character count: since truncation takes the end and the device decides where the end falls, the words that distinguish your page belong at the front. The description should say what the reader will find rather than restate the title in other words. Google sometimes replaces both with its own if it judges them a better fit for the query — that too is normal behaviour, not a fault; where the replacement comes from we set out in SEO copywriting.

In the text we merged, this section carried a promise of a click-through improvement "of up to 40%". It had no source and was removed along with a dozen other numbers of that kind.

Technical work: what Google calls a signal, and what is folklore

One rule applies here: read the documentation, not other people's summaries of it. Two things are worth knowing by heart.

First: Core Web Vitals are a signal, not a passport. On its page experience page Google writes that "Core Web Vitals are used by our ranking systems". In the same document, however, stands the sentence that vanishes from the summaries:

> Google Search always seeks to show the most relevant content, even if the page experience is sub-par.

And, just as plainly, that beyond Core Web Vitals other page experience aspects do not directly help a site rank higher. In other words: relevance beats performance scores. A company paying several thousand for "green numbers in PageSpeed Insights" without content that answers the query is buying speed — valuable to users — but not position. The measurement itself is free and takes a minute; you can run our own speed test before anyone invoices you for it.

Second: one of the three metrics no longer exists. On 12 March 2024 INP (Interaction to Next Paint) replaced FID (First Input Delay) as a Core Web Vital. The announcement on web.dev says FID "will no longer be a Core Web Vital, and will be officially deprecated and removed from the program". That is a convenient date stamp when judging material and proposals: if an audit or a guide written in 2026 still measures FID, it is a copy of a text from before March 2024. Our own article, the one merged here, taught FID until today — which is why we say this about ourselves as well.

The rest of the technical layer — an HTTPS certificate, a sitemap, no mass of 404s, correct redirects when addresses change, working on a phone — is the foundation whose absence hurts and whose presence does not lift. A foundation is not sold as an advantage.

Links: what Google forbids in its own words

Links from other sites are a signal and that has not changed. What has changed is that you can now quote what not to do instead of arguing about it.

The same documentation page that rules out ranking guarantees warns against suppliers offering link popularity schemes and submitting your site to thousands of search engines. Techniques that break Google's rules risk, again quoting, a drop in ranking or removal from the index. Buying links and private blog networks fall inside that description.

The part you control entirely and which costs nothing is internal linking: whether your home page and your articles lead to the pages meant to earn. It is also one cause of the "unknown to Google" state — an address nothing points to is often simply undiscovered. Twelve per cent of our addresses are in that state and that is our fault, not Google's.

Understand it literally: Google reaches pages by following links. A sitemap tells it what exists, but links decide what looks important. A service page reached by one link in the footer gets exactly as much attention as that one link implies, regardless of how important it is to your sales.

Three things that take an hour and cost nothing: link from articles to service pages where the reader is already asking who could do the work; use descriptive anchors, the name of the topic rather than "click here", because the anchor tells Google what the destination is about; and check that every page you care about can be reached from the home page — if you have to use site search to find it, no link points there.

How to read a report without fooling yourself

A search report rests on four numbers that are easy to confuse: impressions (how often your address appeared in results), clicks, average position and the number of queries Google shows you for. Each can mislead on its own, and together they can tell a story that did not happen.

We will show it on our own example, because we fell for it. In one review we saw that several dozen of our addresses had an average position inside the top ten, and that fewer than one impression in a hundred ended in a click. It looked like a finding: so many pages so high and nobody clicking, so all we need is better titles. We checked before changing anything, and the finding dissolved into three ordinary things:

• A brand query split across a dozen addresses. When somebody types a company name, Google shows them the home page, the about page, the offer, contact and resources — in our case one query spread across more than a dozen of our addresses at once, each at roughly position one. They click one, obviously. Counted per address it looks like a dozen pages in the top ten with no clicks.
• A tail of queries you cannot see. One of our pages had an average position of 9.5 and a respectable number of impressions, and the query breakdown showed a single named query with a single impression. The rest sits below the threshold at which Search Console names queries at all. The average position there is an average over things that cannot be inspected.
• Pages that are not meant to collect clicks. Privacy policy, sitemap, contact form. They are indexed, they draw impressions and they should not have traffic.

The conclusion is practical and uncomfortable for both sides of the table: none of those four numbers is a result on its own. The result is how many queries your site appears for, and whether the ones that matter commercially are growing. A report giving positions without queries and impressions without a breakdown is not a report, it is a chart.

Four sentences you will hear, and what the data says

What to ask a supplier

The list below is not our opinion about the market. It is taken from the warning signs Google names in its own documentation, and from things that can be checked in data:

1. Was a ranking guaranteed? If so, the conversation is over — see the quotation at the top.
2. Did the supplier contact you unprompted? Google lists that as a warning sign.
3. Can they say clearly what they will do? Secretiveness is on the same list.
4. How many of our pages are indexed today, and how many sit in "crawled — currently not indexed"? If they do not know the number, they have not opened Search Console.
5. How many pages drew at least one impression last quarter? That is a measure of visibility; the position of a single phrase is not.
6. Where do the phrases come from and does anyone actually search them? "From competitor analysis" is not an answer — competitors are wrong too, as our own example shows.
7. Does the report mix positions with traffic, and traffic with queries? Three different things, usually in one table.

The question of how long this takes has its own answer and its own article — we cover it there, because an honest answer takes more than a paragraph.

What this article deliberately leaves out

Search spills into several subjects, each of which has its own text here. Repeating them would turn this article into a miniature of the whole site — which is exactly what the three texts it was merged from had become:

• local visibility, the Google business profile, reviews → the Google Business Profile section,
• writing content for search → SEO copywriting,
• measuring speed and Core Web Vitals in practice → website testing tools,
• search for online shops → ecommerce SEO,
• what to do with traffic once it arrives → conversion rate.

If you remember one thing: before you pay for positions, check your index status. It is twenty minutes in a free tool and it settles whether there is anything to discuss.

Sources used in this article, all checked directly on Google's own pages in September 2026: "Do you need an SEO?" (ranking guarantees, warning signs), Search Console help (index states), the page experience documentation (Core Web Vitals and ranking) and the metric change announcement on web.dev (INP replaces FID, 12 March 2024). Figures about our own site come from Search Console for the domain digitalvantage.pl: index status as at 9 September 2026, visibility over the period 18 June – 15 September 2026, given as shares.