An SEO strategy is the list of searches you decide to own, the pages that cover them and the number you agree to be judged on. The audit, the keyword research and the link building are not the strategy: they are the activities that carry out those three decisions.
The hard part is not writing the list of activities. It is deciding which searches to give up, and working out which metric to judge the work on when the number of ranking keywords and the share of visibility move in opposite directions. That happens more often than you would think, and below are the numbers from two client projects where it did.
What an SEO strategy is, and what it is not
Semrush, in its guide updated in August 2026, defines it as "a set of business decisions, not a to-do list". The distinction is worth taking literally, because it gives you a quick test: if a strategy document contains nothing you have given up, that document is a list of activities with a different title.
Giving things up is the uncomfortable part. A manufacturer of packaging machinery that sells six-figure lines has no reason to fight a classifieds portal for the search "used packaging machines": it would pick up the query, lose the comparison, and meanwhile fail to cover the six technical searches its buyers actually use to find it. That is the difference from an SEO campaign: the campaign executes, the strategy decides what the campaign will not do. If SEO is still new ground for you, the basic definition is in what is SEO.
The number of keywords on page one tells you little
Macropix is a Milan-based manufacturer of custom LED screens, with more than five thousand installations worldwide. We have worked on the project for seven years and the historical numbers are public, in the case study on SEO for manufacturing: the keyword "monitor pubblicitario" (advertising display) went from position 88 to position 2, and in July 2026 the domain held 25.55% share of voice across the LED wall cluster.
What follows is the latest month of that same dashboard, read on 11 September 2026 across 135 tracked keywords on Google Italy.
| Metric (Macropix project, 135 keywords) | 13 August 2026 | 11 September 2026 | Change |
|---|---|---|---|
| Keywords in the top 3 | 33 | 30 | -9.1% |
| Keywords on page one (top 10) | 89 | 77 | -13.5% |
| Keywords on the first two pages (top 20) | 115 | 102 | -11.3% |
| Domain share of voice | 8.67% | 21.26% | +12.6 points |
Keywords on page one fell by 13.5% and the share of estimated clicks more than doubled. It is not a measurement error: positions dropped on side searches with low volume, while they rose on the searches where demand is concentrated. Over the same period amazon.it stayed first with 26.89%, and macropix.it went from a distant position to second.
The practical point: if the monthly report you receive has a single line and that line is "keywords in the top 10", you are reading a number that can improve while the project gets worse, and the other way round. Always put an estimate of click share next to it, and read the two together. The other signals worth watching are among the ranking factors and the SEO goals you wrote down at the start.
Search volume has become an unstable foundation
The query "seo strategy" in the UK averages 720 searches a month over the last twelve months, according to Google Ads data, but August 2026 recorded 260, the lowest month in the series (DataForSEO, Google UK, September 2026). Its first page now opens with an AI-generated box that answers the question before any organic result.
On the effect of that box, the most solid data is still American. The Pew Research Center tracked the real browsing of 900 US adults across 68,879 Google searches during March 2025: where an AI summary appeared, users clicked a result on 8% of visits, against 15% where there was no summary. Links inside the box got 1%. The summary appeared on 18% of the searches in the sample. It is a US measurement, across desktop and mobile, and there is no equivalent study on a probability sample for the UK yet: assuming it holds here would be a leap.
The consequence for planning is direct. Building a content plan on the estimated volume of informational searches means resting it on a foundation that is moving downwards while you use it. Queries with commercial and transactional intent, the ones where someone is choosing a supplier, shrink much less: that is where the weight should go. We have collected the data on how this changes day-to-day work in how AI Overviews affect SEO.
The decisions to make before opening a tool
Five questions, in this order. None of them needs a subscription to anything.
- Which searches you give up. Write them down. A document with nothing given up has no priorities, only a list.
- What you are credible on. A company that has installed five thousand screens can write about heat dissipation and pixel pitch. It cannot write about business grants, and if it does it spends budget on content that will not stand up to the competition.
- Which pages you already have. There is almost always one that picks up the search and serves it badly. Rewriting it costs less than adding another, and it does not create two URLs that take positions from each other.
- Which number you will be judged on. One, chosen in advance. Enquiries, quote requests, click share on the cluster: whatever you like, as long as it is decided before you see the results.
- When you stop. The threshold beyond which a cluster has given what it can, and resources should move elsewhere.
The fourth question is the one most often skipped. A commitment made in September and checked in March is worth more than twelve monthly reports that change metric every time the previous one is unflattering. Semrush organises the same reasoning into six decisions, from who you want to reach to how you will measure it; Shopify Academy, in its guide, separates the case of a new brand from that of an established one, because in the first case part of the budget has to create the brand demand that already exists in the second. Both guides sit on the first page of Google UK for this query (DataForSEO, September 2026).
The answers to the first and third questions come from keyword research and from reading search intent, which here are used to discard, not to accumulate.
A new site and a site with history do not share a strategy
If the site is already a few years old, the first source is not a third-party tool: it is Google Search Console, which holds the queries you already appear for without having targeted them. Pages with stable impressions and falling clicks are the natural candidates for rewriting, and in 2026 they are also the signal that a search has moved to the AI box. The rest of the diagnosis goes through an SEO audit.
If the site is new or nearly new, the constraint is different: you have no data of your own and have to decide in the dark. It helps to remember where the market starts from. According to the ONS analysis of the Business Insights and Conditions Survey published in July 2026, 35% of UK businesses with ten or more employees used at least one AI technology in June 2026, up from about 12% in September 2023, but only 10% reported using it extensively. Put simply: in many B2B niches competitors have a website and a few tools, not a strategy, and the bar for overtaking them is lower than the guides written for large brands suggest.
A second Italian project we work on, in a different sector from Macropix, shows what happens when you start from a weak base: out of 143 tracked keywords, those on page one went from 5 to 45 in thirty days, and those on the first two pages from 33 to 79. Ninety-five keywords moved up in the same month. The client has not authorised us to publish its name, so the figure stays anonymous, but the order of magnitude is the point: on a site with no history the early movements are wide and misleading, and the first reliable number comes later, not straight away.
In what order to spend resources
The order matters more than the list, because each item unlocks the next.
| Work | When it makes sense to do it first | How you know it worked |
|---|---|---|
| Indexing and technical accessibility | Always, if important pages are not indexed | Indexed pages in Search Console |
| Rewriting pages that already rank | If there are pages with high impressions and low clicks | Clicks and CTR on the same queries |
| New pages for uncovered searches | If current pages do not cover a query with commercial intent | First useful positions within 3-6 months |
| External authority | If you are stuck on page two for contested queries | New referring domains and movement on the stuck queries |
| Presence in generated answers | If queries in your sector already show the AI box | Citations and mentions in AI answers |
The last row carries a misunderstanding worth clearing up. Google's official documentation is explicit: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary", and there is no need to add machine-readable files or dedicated structured data. A page has to be indexed and eligible to appear with a snippet, full stop. Anyone selling a "GEO" service separate from SEO is, on the technical side, selling the same thing twice. The part that does change, how you write and how you measure presence in generated answers, is covered in SEO and AI and in AI content and SEO. For external authority, the criteria are in the link building guide.
How long to wait before judging
At thirty days you judge execution, not results: the pages are live, they are indexed, the technical issues are closed. At three months you see the first movements on less contested queries. At six months you can say whether the approach holds. The Macropix project reached second position on a query with 390 monthly searches in Italy, starting from page nine, but the project is seven years old: anyone promising the same result in a quarter is describing something else. Typical timelines and costs are covered in how long SEO takes and in how much SEO costs.
What is already published matters more than what you will publish
This is the part almost every content plan skips, and it is where growing blogs put a large share of their resources. Backlinko, in its guide updated in April 2026, sorts the work on existing content into three levels of increasing depth (tweaks, upgrades, complete rewrites) and brings numbers from its own domain: a 70.43% increase in organic traffic after updating a single article, and 164,000 clicks in twelve months (+626%) after consolidating pages that were competing with each other. These are figures from a single large US site, so they do not transfer as they are, but the mechanism they describe, consolidating instead of multiplying, works the same way on a smaller site and costs less than producing from scratch.
One day a month set aside only for updates, marked in the calendar, is the practical difference between an archive of articles and a site that holds its ground. The concrete tasks for that day are listed in how to improve your site's SEO.
If you would rather someone else took this part on, it is the work we do in our SEO services, and the results you can hold us to are in our projects.
Frequently asked questions
The strategy document itself (analysis, choice of the searches to target, page plan, definition of the metrics) is a one-off piece of work, and execution is a separate recurring cost. The variable that moves the price most is not the sector but the number of existing pages that need fixing: a small business site and an e-commerce site with thousands of URLs are different jobs.
They fall into three families: technical SEO (making sure Google can read and index the pages), on-page SEO (making a page the best answer to a specific search) and off-page SEO (earning mentions and links from other sites). They do different jobs and should be switched on in order: optimising the text of a page Google cannot index achieves nothing.
The decision-making part, yes, and it is better that way: no consultant knows the margins per product and the customers you really want better than the people running the business. The parts where outside expertise helps are reading the SERP (understanding why one page ranks and another does not) and technical diagnosis. If you are starting from zero, begin with Google Search Console, which is free and already holds your own site's data.
Fewer than you will find. On a mid-sized B2B project the searches that bring qualified enquiries are often around ten, and the rest of the traffic is noise. A large number of tracked keywords is useful for monitoring, not for planning: the right question is how many searches you want to hold steadily, not how many you can brush against.
On the technical side, no. Google's documentation says there are no additional requirements or special structured data to appear in AI Overviews or AI Mode: the page has to be indexed and eligible to show a snippet. What changes is measurement, because a citation inside a generated answer does not show up as a click in traffic reports, and has to be tracked separately.
One thing the industry rarely says: the most useful part of a strategy document is the list of searches you decided to leave to others, and hardly anyone writes it. Without that list, twelve months later nobody can tell whether the plan worked or was simply rewritten along the way.
