Improving a website's SEO means increasing the clicks that come from organic search. Not the rankings: the clicks. The distinction sounds pedantic, but it decides the order in which things are worth doing, and before that whether it is worth investing in at all.
An example from real data. On an Italian sports equipment retailer we work with, between 1 May and 25 August 2026 Search Console recorded 45,427 impressions and 1,610 clicks across a thousand queries from the Italian market. Seven of those thousand queries contained the shop's name. They account for 0.7% of the queries and 61% of the clicks.
Where organic clicks really go
This is the full table. The brand row and the generic query rows belong to the same site, over the same period, in the same market.
| Query type | Queries | Impressions | Clicks | CTR |
|---|---|---|---|---|
| Brand queries | 7 | 2,639 | 983 | 37.25% |
| Non-brand, average position 1 | 83 | 1,262 | 92 | 7.29% |
| Non-brand, average position 2-3 | 186 | 12,615 | 300 | 2.38% |
| Non-brand, average position 4-10 | 366 | 19,837 | 205 | 1.03% |
| Non-brand, average position 11-20 | 251 | 8,276 | 27 | 0.33% |
Three things stand out. First: on generic queries, reaching page one is not enough, because between positions 4 and 10 the CTR is 1.03%. Second: the real jump is between second or third place and first, where CTR goes from 2.38% to 7.29%. Third: the brand brings in more clicks than everything else put together, with 0.7% of the queries.
The twelve actions below are ordered on that basis: first the ones that move clicks within a few weeks on pages you already have, then the ones that move them over a few months, and last the ones almost everyone does first and which, on their own, move little. If you are short of time, the first four are worth more than the other eight.
1. Update the pages that are losing clicks before writing new ones
In 2025 Ahrefs repeated a study it first ran in 2017, on two million pages published in English and 1.3 million US keywords. Of the new pages, 6.11% reached the top 10 within twelve months. In the top 10, 72.9% of pages are more than three years old, against 59% in 2017, and the page in first position is on average five years old. The direction is clear: Google's top results are getting older, and a new page starts with a handicap.
The practical way to use this: open the Performance report in Search Console, compare the last three months with the same quarter a year earlier and sort by difference in clicks. Pages with stable impressions and falling clicks are the ones to rewrite first, because Google is still showing them and it is users who are not clicking. If you have not set up the comparison yet, the steps are in how to check a website's traffic.
2. Go after the queries in positions 8-20
In the table above, the 251 queries between positions 11 and 20 produced 8,276 impressions and 27 clicks. The same queries moved into the 4-10 band would be worth about 85 clicks at that band's average CTR. Moved into second or third place, they would be worth almost 200.
These are the queries for which Google already considers you relevant, just not relevant enough. Filter the Search Console query report for position between 8 and 20, sort by impressions and check which page Google is showing for each. In half the cases the page exists but covers the topic in three lines inside another piece: what it needs is a dedicated section, not a new article. How to decide between the two is covered in the piece on keyword research that starts from the data you already have.
3. Rewrite the titles and snippets of pages with many impressions and few clicks
This is the action with the fastest return, because it does not require gaining a single position. Take the pages with more than a thousand impressions and a CTR below the average for their position band, and rewrite the title and meta description after looking at how the other results for that query present themselves.
Two caveats. Google truncates snippets by pixel width, not by character count, so check them on the real results page, not in a counter. And Google rewrites the title when it thinks yours does not describe the page, which happens more often with titles stuffed with keywords. The practical details are in meta title and meta description; the metric to track is explained in what CTR means in SEO.
4. Match the page to the intent of the results page, not your own
If you search for your keyword and see ten product pages, an article will not rank, however well written it is. The reverse is also true: many service pages do not rank because the results for that query are made up of guides.
The check takes two minutes per keyword and should be done before rewriting anything: search the query in an incognito window, look at what type of page holds the top five places and adapt the format. If the format does not fit, change keyword. The four types of intent and how to recognise them from the results page are in search intent.
5. Point internal links at the pages that need to climb
It is the only authority lever you control completely. Search Google for site:yoursite.co.uk "page topic", open the first ten results and from each one add a link to the target page inside a sentence that already talks about that subject. Descriptive anchor text, never inside a heading.
On our blog this step has become a fixed rule: every new article gets three to five links from pieces already published, on the day it goes live. How it works is explained in internal links, and the comparison with external links in link building.
6. Merge pages that compete for the same query
Filter the Search Console query report on a specific query and add the Page dimension. If two or three URLs appear with split impressions, Google is alternating between your pages and none of them builds up signals. The fix is almost always a single, more complete page and a 301 redirect from the others.
It happens a lot on sites that have published for years without a map: articles on neighbouring topics that half overlap. A crawler such as Screaming Frog or Sitebulb helps find them by comparing similar titles and H1s, but the confirmation always comes from Search Console. The difference between this and true duplicate content is explained in duplicate content.
7. Answer long questions in a dedicated section
The Pew Research Center analysed 68,879 Google searches made by 900 US adults in March 2025. Searches of one or two words produced an AI summary 8% of the time; those of ten words or more 53% of the time; those phrased as a question 60% of the time. It is a US sample, but the logic of the query is the same everywhere: the longer and more detailed the question, the more likely Google is to answer it itself.
The practical consequence is not to stop covering those questions but to cover them in a format that can be cited: the question as a heading, a plain answer in the first two lines, detail underneath. It is the same pattern that wins featured snippets, and the one answer engines are most willing to lift, as we explain in how to rank in answer engines.
8. Grow the number of people who search for you by name
Back to the opening table: 37.25% CTR on brand queries, 1.47% on all the others combined. A twenty-five-fold multiplier that no on-page optimisation can come close to.
In practice, activities usually seen as outside SEO, from newsletters to trade shows, from video to partnerships, feed the organic channel more than many technical optimisations. The way to check this in your own data is to split Search Console queries into those containing the company name and all the rest, and watch how the two curves move over time. If brand grows and non-brand does not, the problem is content; if non-brand grows and brand stays flat, you are attracting visits from people who do not remember you.
9. Work on performance only if you are over the thresholds
The official Core Web Vitals thresholds are known and public: LCP within 2.5 seconds, INP within 200 milliseconds, CLS within 0.1, measured at the 75th percentile of real visits. Below those thresholds you are fine. Above 4 seconds, 500 milliseconds and 0.25 the page is classed as poor, and that is the band where spending money makes sense.
Google is explicit about how much these numbers weigh: its page experience documentation says there is no single signal and that good scores do not guarantee high rankings. The numbers to look at are real-user data, collected in the Chrome UX Report and shown by PageSpeed Insights in the box at the top, not the lab score below it. Going from 2.8 to 2.1 seconds on a site that is already decent moves little; going from 6 seconds to 3 moves a lot, especially on conversion rate. When it is worth tackling is the subject of page speed.
10. Add structured data where it changes the snippet
Structured data does not make a page climb. It changes how the page looks, and that changes CTR. It makes sense for products with price and availability, recipes, events, reviews with stars, FAQs and articles with an author. On a generic service page it produces nothing visible.
The rule for not wasting time: implement only the types for which Google documents a rich result, then run the page through the Rich Results Test and check that the markup is eligible, not merely valid. A plugin such as Yoast already covers article, breadcrumb and organisation: everything else has to be added by hand. The full picture of what it does and does not do is in schema markup.
11. Get cited outside your own site
Again from the Pew data: when an AI summary appears, 8% of visits end with a click on a traditional link, against 15% when there is no summary; 1% click a link inside the summary; 88% of summaries cite three or more sources. With a summary, 26% of sessions end there, against 16% without. US data, March 2025.
We can already measure the effect on our own clients: on the keyword cluster of an Italian industrial client, 32.7% of keywords show an AI Overview, equal to 24.7% of total search volume, and when the AI Overview appears 43% of searches end without any click. If 88% of summaries cite three or more sources, being one of those sources is the position that counts. How to work on this is explained in AI Overviews and in our SEO for generative engines service.
12. Measure money, not rankings
An average position that improves from 14 to 9 means nothing to a sales director. A cost per lead does. The fairest benchmark is the paid channel on the same audience: on a Google Ads account we manage, between July 2025 and June 2026 there were 374 conversions from €16,805 spent, or €44.93 per conversion, with an average CTR of 3.59% and an average CPC of €0.81. If that comparison does not add up, the problem is usually not ranking but the point where the site stops converting.
That number is the yardstick. If SEO on those same queries costs less than €44.93 per lead, it is worth pushing; if it costs more, it is better to move budget. It is the same reasoning we apply in our projects and the basis of our SEO service.
Five things Google says not to do
Half the time spent improving a website's SEO goes on practices that the official documentation has contradicted for years. This table summarises what Google says, word for word, in its starter guide.
| Practice | What Google says |
|---|---|
| Filling in the keywords meta tag | Google Search doesn't use the keywords meta tag |
| Managing keyword density | Repeating the same words is keyword stuffing and is against the spam policies |
| Meeting a minimum word count | The length of the content alone doesn't matter for ranking purposes |
| Following a strict heading hierarchy | Having your headings in semantic order isn't relevant for Search |
| Optimising for E-E-A-T as a ranking factor | E-E-A-T is not a ranking factor |
The last row surprises people who are new to SEO. E-E-A-T is still the vocabulary Google's quality raters use to assess pages and it is still useful as a writing standard, but it is not a dial to turn: the full explanation is in E-E-A-T, and the picture of what really matters in ranking factors.
How long before you see results
Google says some changes take effect within a few hours and others take several months. The Ahrefs study adds a useful detail: of the pages that eventually reach the top 10, 40.82% get there within the first month. Put another way, if a page does not move in the first thirty days, the problem is almost never patience but content.
For rewrites of pages that are already indexed the timescales are shorter: titles and snippets show up within days, reworked content within two or three weeks. Estimates by type of action are in how long SEO takes to show results.
A case study, with numbers
Macropix makes custom LED screens in Milan. In 2020 it ranked 88th for the keyword "monitor pubblicitario" (advertising monitor), on the ninth page of Google Italy. In 2025 it is in position 2. For "ledwall", with 4,400 searches a month in Italy, it went from absent to first. In July 2026 its share of voice on the ledwall cluster, measured with Semrush, is 25.55%, ahead of amazon.it at 18.08% and hddsvision.it at 10.17%.
It did not happen thanks to a technical fix. It happened because for years the work went into the same pages instead of adding new ones, and because the catalogue was reorganised around how people search, not how the company names its products. The full story is in SEO for manufacturers.
One last reference point for how much room there is in that market. According to the 2025 Istat survey on businesses and ICT, in Italy 14.7% of companies with at least ten employees sell online and 38.1% reach a high level of digitalisation. In most Italian sectors organic competition is still thin, which is part of why a manufacturer could climb from page nine to position two.
The metric missing from almost every dashboard
The ratio between brand and non-brand clicks is the first thing to look at, and hardly anyone tracks it. If the brand makes up 61% of clicks, as in the opening case, generic SEO is still a small channel, with large room for growth and low risk: whatever you do, you are unlikely to make things worse. If the brand makes up 10%, you are already exposed to competitors on every single query, and every point of CTR gained is worth much more than a position.
That ratio takes ten minutes to calculate with a filter in Search Console, and it changes which of the twelve actions on this list you should do first. If you have never measured it, do it before you open the editor.
Frequently asked questions about improving SEO
It depends on the type of action. Rewriting the title and meta description of a page that is already indexed has an effect within a few days. Reworking the content of an existing page shows within two or three weeks. A new page takes much longer: according to the 2025 Ahrefs study only 6.11% of published pages reach the top 10 within twelve months, and of those that get there, 40.82% do so within the first month.
In most cases, update. In Google's top 10, 72.9% of pages are more than three years old and the page in first position is on average five years old (Ahrefs 2025, English-language sample). Pages that have already built up impressions start with an advantage over a newly published page.
It matters if you are over the thresholds. The official Core Web Vitals thresholds are LCP within 2.5 seconds, INP within 200 milliseconds and CLS within 0.1, measured at the 75th percentile of real visits. Google states that there is no single page experience signal and that good scores do not guarantee high rankings. Going from 6 seconds to 3 changes something; going from 2.8 to 2.1 almost nothing.
Yes, where they appear. The Pew Research Center, analysing 68,879 searches by 900 US adults in March 2025, measured clicks on a traditional link in 8% of visits with an AI summary against 15% without. On a cluster of Italian industrial keywords we monitor, 32.7% of keywords show an AI Overview, and in those cases 43% of searches end without a click.
There is no minimum. Google Search Central's starter guide says, word for word, that the length of the content alone doesn't matter for ranking purposes. The useful reference is not a word count but the results page: if the top five results answer in eight hundred words, writing three thousand will not help.
