The verdict
Google indexes and ranks the mobile version of your pages. Anything a template drops on phones, a hidden section, a missing table, different structured data, a shorter title, does not exist for Google. The check takes ten minutes: compare what a phone and a desktop actually render, starting with the pages that earn money.
When a page I care about is not being indexed, the first thing I do is open it on a laptop and on a phone, in a browser, as a customer would. A surprising number of problems end right there, because the page is visibly not what somebody believed they had published.
The phone half of that check matters more, and the reason is in one sentence of Google’s documentation.
Its mobile-first indexing guidance: “Google uses the mobile version of a site’s content, crawled with the smartphone agent, for indexing and ranking.” And, more bluntly: “Only the content shown on the mobile site is used for indexing.”
The same page asks for the mobile and desktop versions to carry the same content, the same structured data, and an equivalent title and meta description. It warns against lazy-loading primary content “upon user interaction”, because Googlebot does not tap, swipe or type to reveal it.
Tabs and accordions are fine. Google suggests them as a way to keep mobile pages compact without cutting anything. Hidden is not the same as missing.
Page builders are common enough for this to be the normal case. The same Almanac found about 60% of WordPress sites use a page builder, with Elementor alone on 43%.
Take the five pages that earn money. Open each on a phone and a desktop, and compare section by section: headings, body copy, tables, reviews, FAQs. Then use the URL Inspection tool in Search Console on the same page and look at what the smartphone crawler rendered. If something is on the desktop and not in that render, Google does not know it exists.
Mobile parity is one of the causes behind a page being crawled and passed over. The rest are in why pages get crawled and not indexed.
If a site should be ranking and it isn’t, that’s the work I do. Technical SEO covers the rendering and indexing layer, which is where this lives.
]]>The verdict
Google calls a sitemap a hint, ignores its priority and change-frequency tags, and only trusts lastmod when it is consistently accurate. So a sitemap will not make Google crawl or index anything. What it does well is diagnosis: compare what you submitted with what was indexed, template by template, and the gap tells you where to look.
Sitemaps get treated in one of two ways. Either somebody submitted one years ago and considers that part of SEO done, or nobody ever submitted one at all. Both treat it as a switch. It is more useful than that, and less powerful.
Google’s sitemap documentation, updated in July 2026, is unusually blunt about the limits. Submitting one “is merely a hint”, and it “doesn’t guarantee that Google will download the sitemap or use the sitemap for crawling URLs.”
Most of the optional tags do nothing. “Google ignores <priority> and <changefreq> values.” The last-modified date is used only “if it’s consistently and verifiably” accurate, and it should reflect the last significant change to the page. Changing the copyright year in the footer is not one.
The hard limits are 50,000 URLs or 50MB per file. Anything larger is split into several files under a sitemap index.
It cannot rescue an orphan page. A classic car marketplace I audited had expired listings that stayed in its sitemap indefinitely, with nothing on the site linking to them. Google knew every one of those URLs existed, because the sitemap told it so. Most of them still went nowhere. Discovery is not the same as being worth indexing, and a sitemap only does discovery.
The split by template is a habit of mine rather than a Google requirement. It costs nothing on most platforms and it turns one vague number into several specific ones.
The sitemap is one of the technical causes behind pages Google fetches and then declines. The full order I work through is in why pages get crawled and not indexed.
If a site should be ranking and it isn’t, that’s the work I do. Technical SEO covers crawling, indexing and the sitemap that is supposed to describe them.
]]>The verdict
Changing the date on a page is not an update, and Google’s own self-assessment asks about exactly that. A refresh earns its time when a page is losing impressions for queries it used to win, or when what it says has gone out of date. Change the substance, then the date, and label which date is which.
A common piece of advice says to update old posts regularly because Google likes fresh content. The common way of following it is to change the date, tweak the introduction and republish. It takes ten minutes and it is very nearly the opposite of what Google asks for.
Google’s helpful content guidance includes a self-assessment question aimed at exactly this habit: “Are you changing the date of pages to make them seem fresh when the content has not substantially changed?” It sits in a list of warning signs of content made for search engines rather than people.
Its guidance on dates treats published and updated as two different things, and asks for them to be labelled clearly. A page can show both.
Google does run freshness systems. Its ranking systems guide describes them as showing newer content where newer content is expected, with examples like a newly released film or a recent earthquake. That is a property of the query, not a reward for every page that changes its date. “How to fix a leaking tap” does not get better because it was published last week.
A section that answers what the query is asking now. Facts corrected. Things that were wrong removed, and ideally said so. When I changed my mind on two recommendations I had been making in audits for years, I wrote a post saying which two and why, rather than quietly editing the old advice. That is the standard I would apply to a page as well: if it changed, say what changed.
Then update the date, labelled as an update, alongside the original publication date.
A slow loss of impressions is one of the patterns behind a high-impression, low-click row. The others are in four causes of high impressions and low clicks.
If a site should be ranking and it isn’t, that’s the work I do. An SEO audit shows which pages are slipping and why, so the refresh goes where it will move something.
]]>The verdict
Keyword density mattered in the 1990s, when search engines counted words. Google’s starter guide now says excessively repeating the same words is tiring for users, and keyword stuffing is a named spam policy. Tools still score density. I ignore the score and use the page map: the right words in the title, the H1 and the opening, then write for the reader.
Plenty of content briefs still ask for the main keyword at a set density, often somewhere between one and two per cent. Follow one closely and the page reads like it was written by somebody trying to remember a phone number.
The rule is older than Google, and tools are still printing it.
The earliest search engines relied heavily on what a page said about itself: the words on it, how often they appeared, and the keywords meta tag in the page head. Wikipedia’s history of search engine optimization describes how quickly webmasters learned to manipulate exactly those signals, which is a large part of why link-based ranking replaced them.
By 2012 Google was writing about keyword stuffing as webspam. Its April 2012 announcement of what became the Penguin update named “keyword stuffing” alongside link schemes, and said the change affected about 3.1% of English queries.
Its SEO starter guide is plain on three points. “Google Search doesn’t use the keywords meta tag.” “Excessively repeating the same words over and over (even in variations) is tiring for users.” And “The length of the content alone doesn’t matter for ranking purposes.”
Keyword stuffing is listed in Google’s spam policies, and “Hidden text and/or keyword stuffing” is one of the manual actions Google can apply. On local sites the usual version is a footer listing forty suburbs.
Because it is easy to measure. A tool can count words. It cannot tell whether a page answers the question better than the ten above it. So the score survives because it produces a number, not because the number means anything.
Where the words go is decided by the keyword map, not by a percentage. That process is in how I build a keyword map for a site that already exists.
If a site should be ranking and it isn’t, that’s the work I do. An SEO audit checks what each page is actually telling Google it is about, which is a better question than how often it says so.
]]>The verdict
Search Console shows which queries brought clicks, not which brought customers. A small, tightly limited paid search test does. The search terms report in Google Ads lists the actual searches behind each enquiry, which is the best keyword research a new page can have. It costs money, so I keep the test small and the question narrow.
Every keyword tool can tell you how many people search for something. None of them can tell you which of those searches turn into a phone call for your business. That is the question that decides whether a page is worth building, and there is one place that answers it directly.
It shows impressions, clicks and positions per query. It does not show which query produced an enquiry, because it has no idea what happened after the click. Its tables also stop at 1,000 rows, and it only contains queries where you already appear. For a page that does not exist yet, it has nothing to say at all.
Google Ads has a report of the actual searches that triggered your ads. Google describes it as showing how ads performed when “triggered by actual searches”. Join that to conversion tracking and you can see, query by query, which searches turned into calls and forms.
Two limits are worth knowing. Google omits terms without enough query activity, for privacy. And it only reports searches your ads were eligible for, so the test is only as good as the terms you put into it.
When the page already ranks and Search Console has months of data on it. When the clicks are so expensive that a meaningful test costs more than the page is worth. And when nobody will answer the phone during the test, because then the data measures your intake, not the keywords.
This is a judgement I would make per business rather than a rule. It is also the one part of keyword research that tells you what a search is worth, rather than how often it happens. The rest of the process is in how I build a keyword map for a site that already exists.
If a site should be ranking and it isn’t, that’s the work I do. An SEO audit tells you which pages are worth building before anyone spends on them.
]]>The verdict
Two pages appearing for one query is not automatically cannibalisation. It becomes a problem when they take turns: one page this week, the other next week, and neither settles. Search Console shows it if you filter by the query and look at pages over time. The fix is usually deciding which page owns it, not deleting one.
Keyword cannibalisation is one of the findings audit tools report most generously. Any two pages that rank for the same word get flagged, and a site with a few hundred pages can come back with a list of hundreds of “conflicts”.
Most of them are nothing. A few of them are costing real traffic. The difference is visible in Search Console in about ten minutes.
Real cannibalisation has a shape. Two of your pages appear for the same query in the same period, at similar positions, and they take turns. One week the service page shows, the next week a blog post does, and the average position for the query jumps around because Google keeps changing its mind about which of your pages is the answer.
Google generally avoids showing several near-identical pages for one search. Its ranking systems guide describes deduplication systems that keep only the most relevant of very similar results. When two of your pages are close enough to compete, that is the system choosing between them, and it may not choose the one you would.
When the two pages answer different intents. A service page and a blog post can both appear for a broad phrase because one serves someone ready to hire and the other someone trying to understand the problem. Sometimes Google shows both, and two results from one site on the same page is not a problem anybody needs to solve.
Most cannibalisation is prevented, not fixed. The rule that prevents it is in how I build a keyword map for a site that already exists: one keyword, one page, decided before anything is written.
If a site should be ranking and it isn’t, that’s the work I do. An SEO audit separates the conflicts worth fixing from the ones a tool flagged by default.
]]>The verdict
On a site that already exists, keyword research is mostly a mapping job: which page should own which query, and which queries have no page at all. When I mapped this site, 9,470 keywords went across 19 pages and no keyword was assigned twice. The tool export is raw material. The map is the research.
Most of the keyword research I am handed is a spreadsheet of a couple of thousand keywords, sorted by search volume, with no column saying which page each one belongs to.
That is a list, not research. A keyword with no page attached is not a decision about anything. It is a reminder that people search.
Because nobody can act on it. The writer does not know which page to change. The developer does not know which template matters. The client does not know whether the site needs ten new pages or none.
And the web is already full of pages that were never aimed at a query anybody types. Ahrefs examined 14 billion pages and found 96.55% of them get no traffic from Google. A keyword list that does not say which page should earn which query is how a site adds to that pile.
This is how I did it for this site, which has a home page, a set of service pages and a set of industry pages: nineteen pages that are meant to earn something.
Second, not first. Once the map exists, Search Console tells you, page by page, which queries each one already earns impressions for. Three things come out of that comparison: queries landing on the wrong page, demand with no page at all, and pages earning nothing for anything on their tab.
It cannot be the first pass on its own, because it only contains queries where you already appeared, and Google’s Performance report tables stop at 1,000 rows. It shows your slice of the market, not the market.
The first of those three findings, two pages competing for one query, is common enough that I have written up how to tell real cannibalisation from two pages that are fine separately.
If a site should be ranking and it isn’t, that’s the work I do. An SEO audit includes the page map: which page should own which search, and where the gaps are.
]]>The verdict
StatCounter puts Google at 85.19% of US search in September 2026 and Bing at 9.89%. For most local and ecommerce sites, Bing is worth one afternoon: import the site into Bing Webmaster Tools, check what it flags, claim Bing Places if you serve customers locally, then go back to Google. It is not worth a separate strategy.
Nearly every conversation I have about search is a conversation about Google, and for good reason. But “search” and “Google” are not the same word, and the gap between them is large enough to be worth a deliberate decision rather than a default.
StatCounter’s US search engine market share for September 2026, across all devices:
Shares like these vary a lot by country and over time. Google’s share in Germany was already put at 85 to 90% in 2006. In the US, roughly one search in seven happens somewhere else.
This is a judgement rather than a measurement. I would give Bing more attention where buyers search from work computers on default settings, because Microsoft’s browser and Windows search both default to Bing. That describes a lot of B2B buying and a fair amount of professional services.
For a local trades business or a Shopify store selling to consumers on phones, I would expect Bing’s share of their traffic to sit at or below the national figure.
There is also a newer reason to look. Bing Webmaster Tools now has an AI performance report showing which of your pages were cited in Copilot and Bing’s AI answers, along with the queries used to find them. Google’s equivalent report shows impressions but no queries, so Bing’s is the only first-party view of what an AI system searched for when it cited you.
Write content for Bing. Buy a Bing-specific tool. Run a separate Bing strategy for a site that has not got its Google basics right. The things that make a page rank in one engine mostly make it rank in the other, and the afternoon above is the whole of the difference for most sites.
Market share is one of the facts about search that sits outside Google’s own documentation. What Google itself has put on the record is in what the antitrust trial and the leak put on the record about ranking.
If a site should be ranking and it isn’t, that’s the work I do. If you want to know whether any search engine other than Google deserves your time, that’s a short conversation in SEO consulting.
]]>The verdict
Google’s quality raters score search results to test its systems. Google says their ratings do not directly affect ranking and raters have no control over how pages rank. The guidelines are still worth reading, because they describe what Google is trying to reward. I use them as a trust checklist in audits, never as a list of ranking factors.
There is a steady supply of articles explaining how to “optimise for the quality raters”. The premise is that a person at Google will look at your site and mark it down. That is not what raters do, and Google has said so in writing more than once.
Google works with external raters who score search results against a published set of guidelines. Google’s explanation of its testing is direct about what the scores are for: “These ratings do not directly impact ranking, but they do help us benchmark the quality of our results.”
Its guidance on helpful content, updated in October 2026, goes further: “Search raters have no control over how pages rank,” and “Rater data is not used directly in our ranking algorithms.”
So the raters grade Google, not you. If a change to the ranking systems makes the results worse by the guidelines’ standard, the raters’ scores show it.
Because they are the closest thing Google publishes to a description of what it is trying to reward. Google first published the full document in November 2015, when it ran to 160 pages, and it has been revised many times since.
The same helpful content guidance sets out the idea underneath it, experience, expertise, authoritativeness and trust, and adds two lines worth knowing. “While E-E-A-T itself isn’t a specific ranking factor”, and of the four, “trust is most important.”
Treat “add E-E-A-T” as a task. An author box with a stock photo and a paragraph of adjectives is not trust. It is decoration that reads as filler, because it is.
And never invent it. Made-up credentials, a reviewer who never reviewed anything, a “medically reviewed” badge on a page no clinician read. On the pages where trust matters most, that is the fastest way to lose it with the people you are trying to reach, whatever any system does about it.
The raters are one of several things about how Google judges pages that are now on the record. The rest is in what the antitrust trial and the leak put on the record about ranking.
If a site should be ranking and it isn’t, that’s the work I do. If you want a second opinion on whether a site looks trustworthy to the people deciding whether to call you, that’s part of SEO consulting.
]]>The verdict
Google now says it uses aggregated interaction data to judge relevance, and trial testimony described Navboost adjusting rankings with click data. The conclusion people draw is to chase click-through. Mine is narrower: clicks reward the page that answers the query, a title rarely fixes a mismatch, and small sites have too little click data for it to be the lever.
The question “does Google use clicks?” now has an answer, and it is yes. The more useful question is what follows from that, and the popular answer is mostly wrong.
Google’s own explanation of how it ranks results says it: “We also use aggregated and anonymized interaction data to assess whether search results are relevant to queries.”
The antitrust trial filled in some detail. A former Google engineer described Navboost as “a memorization system” that records which results people clicked for which queries, and as a ranking component that “uses that data to adjust the scores of search results when it’s available.” The transcript is public, and worth twenty minutes of anybody’s time.
Click bots. Services that send fake searches and fake clicks. Title formulas engineered for curiosity rather than accuracy.
The first two are named in Google’s spam policies: “sending automated queries to Google” is machine-generated traffic, and it violates both the policies and Google’s terms. The third is legal, and it mostly does not work, for a reason the testimony itself explains.
Match the query. If a page ranks for searches its customers would not type, it will collect impressions and very few clicks, and the click data that builds up will say it is a poor answer. Sometimes the right move is to stop chasing that query rather than to rewrite the title.
Accept that fewer clicks are on offer. SparkToro’s analysis of Similarweb’s US panel found 68.01% of US Google searches ended without a click to the open web in January to April 2026. The clicks that remain are more deliberate, which makes earning them by being the right answer more important, not less.
And measure the click you care about. For a local business that is the call, the booking, the form. Click-through rate is a symptom. It is not the goal.
This is one of the things the trial confirmed. The rest, and what it did not change, is in what the antitrust trial and the leak put on the record about ranking.
If a site should be ranking and it isn’t, that’s the work I do. If you want somebody to read your Search Console data and tell you which low-click rows are real problems, that’s part of SEO consulting.
]]>