SEO that compounds instead of spiking

Rankings are a proxy, and an increasingly poor one

There was a period when a ranking report was a reasonable summary of an SEO programme. That period is over. Results now vary by location, device, search history and the shape of the results page itself, so a screenshot showing position three describes one query, from one place, on one day. It is a directional instrument, not a measurement, and an agency that reports it as the headline is choosing the number most likely to look good.

The numbers that survive scrutiny are less flattering and more useful: impressions and clicks for the queries you decided to target, segmented by page group so that commercial pages are visible separately from the blog; the proportion of your important URLs actually indexed; assisted and last-click conversions from organic; and revenue where the data supports attributing it. Those tell you whether the programme is working. Rankings tell you how a tool sampled a search engine this morning.

This matters because SEO is the acquisition channel where mistakes take longest to become visible. Paid media punishes a bad decision within its own reporting cycle. Organic search will let you publish for a year against a flawed keyword map before the absence of results becomes undeniable. Almost everything we do differently comes from that asymmetry: decide carefully, structure properly, and measure something that would tell you early if you were wrong.

The technical foundations, and the ones you probably do not need

Technical SEO exists to remove the reasons a search engine cannot see, understand or trust your pages. It is the part of the discipline with the clearest right answers, which is also why it attracts the most inflated invoices.

Start with the questions that actually matter. Can the pages be crawled? Are they being indexed, and if not, what does Search Console give as the reason? Does the content render without JavaScript executing successfully, and if it does not, has anyone confirmed Google renders it anyway? Is one canonical URL declared per piece of content, and do the signals agree? Google is explicit about the hierarchy here: redirects and rel="canonical" annotations are strong signals, inclusion in a sitemap is a weak one, and robots.txt is not a canonicalisation tool at all — blocking a page prevents Google from seeing the canonical signal rather than resolving the duplication.

Sitemaps are worth getting right and easy to get wrong. A single sitemap file is limited to 50MB uncompressed or 50,000 URLs, beyond which you split and use an index file. More usefully, a sitemap should contain the canonical, indexable URLs you actually want ranked. A sitemap full of redirects, parameter variants and noindexed pages is a discovery problem you built yourself.

Speed belongs here too, though the framing is often wrong. Core Web Vitals — Largest Contentful Paint within 2.5 seconds, Interaction to Next Paint at 200 milliseconds or less, and Cumulative Layout Shift of 0.1 or less, each assessed at the 75th percentile of real visits — are a page experience signal, not a ranking lever you can pull to overtake a better page. They are worth fixing because slow pages lose people, and worth fixing at the level where they are actually caused, which is usually the build rather than a plugin. That is web development work as often as it is SEO work.

Crawl budget is probably not your problem

Here is the thing agencies are least likely to tell you: most sites do not have a crawl budget problem. Google’s own guidance scopes the topic to large sites with more than a million unique pages and moderately frequent changes, medium and larger sites of more than ten thousand pages changing daily, or sites where Search Console reports a large share of URLs as discovered but not currently indexed. Below that, crawl capacity is not your constraint, and a proposal that leads with crawl budget optimisation for a two-hundred-page brochure site is selling complexity.

When it genuinely applies — large ecommerce catalogues with faceted navigation, in particular — the work is real and log file analysis earns its place, because it shows what the crawler actually did rather than what a tool predicts it would do. The deeper technical programme lives on our technical SEO page.

Keyword research is intent research

The unit of useful research is not a keyword. It is a person with a problem at a particular stage of solving it.

Two queries with similar volume can require completely different pages: one wants a definition, one wants a comparison, one wants to buy. Grouping by string similarity produces pages that serve nobody in particular; grouping by intent produces pages that answer one question properly. The quickest sanity check is to search the query yourself and look at what is ranking. If the results are all long explanatory guides, a product page will not win that query no matter how well optimised, and vice versa. The results page is telling you what Google has concluded the intent is, and arguing with it is expensive.

The second discipline is mapping: every query group is assigned to exactly one URL. This is the single most valuable output of the research phase, because it prevents cannibalisation — two of your own pages competing for the same query, splitting signals, and each ranking worse than a consolidated page would. We find this on most established sites, usually where a blog has grown alongside a service section without either being aware of the other.

The third is honesty about competitiveness. Some queries are not currently winnable for your domain, and the correct response is to say so and target the ones adjacent to them. A plan that includes the head term of your industry on a domain with no authority is a plan that produces nothing for a year and then requires an explanation.

Structure: the part that gets skipped

Topical authority is a real effect with an unglamorous mechanism: covering a subject comprehensively and linking those pages to each other coherently makes it much clearer what your site is about. It is not a score, and no tool measures it, whatever a dashboard claims.

In practice this means a hub page owning the broad topic and supporting pages owning specific sub-questions, all linked deliberately rather than by whatever the CMS decided to put in a sidebar. Internal linking is the most underused instrument in most programmes: it is entirely within your control, costs nothing, and directs both crawlers and authority towards the pages that make money. Descriptive anchor text, links placed in body content where they are genuinely relevant, and a deliberate flow from informational pages towards commercial ones will outperform a great deal of external link building.

Structure is also where migrations go wrong. If URLs change, every existing URL needs a mapped destination before anything moves, and the pages currently earning traffic need to be identified beforehand so that nobody quietly deletes one during the content rewrite. Traffic lost in a migration is rarely recovered in full, and “it will come back” is not a plan.

On-page work, minus the folklore

On-page optimisation has accumulated more superstition than any other part of the discipline, largely because it is the part people can do without permission from a developer. A few things are worth being precise about.

The title element still does real work, but not the work most people assume. It is the strongest single on-page statement of what a page is about, and it is also the line a person reads before deciding whether to click — those two jobs sometimes conflict, and the second one is usually more valuable. It is also frequently rewritten by Google when the original does not describe the page well, which is a useful diagnostic: if your titles are being replaced, the page and its title probably disagree.

Meta descriptions do not affect ranking. They affect whether someone clicks, and they too are regularly rewritten in favour of a passage from the page that better matches the query. Write them as an argument for the click rather than as a keyword container, and accept that Google will sometimes overrule you.

Headings should describe the structure of the content, in order, without skipping levels. The version of this advice that says to put your keyword in an H2 has it backwards: a well-organised page tends to contain its terminology naturally, and a page that needs keywords inserted into headings usually has a content problem rather than a markup one. The same applies to image alt text, which exists so that a page is usable by someone using a screen reader and only incidentally helps a search engine — writing it for the second purpose produces text that fails the first.

What actually moves on-page performance, in our experience of auditing sites that have already done the basics, is more structural: whether the page answers the query in its opening rather than after four paragraphs of preamble, whether the specific question a searcher had is addressed explicitly enough to be extracted as a snippet, and whether the page links onward to the things a reader will want next.

Content: briefs, standards and the refresh cycle

Google describes its systems as trying to identify content that demonstrates experience, expertise, authoritativeness and trustworthiness — E-E-A-T — with trust weighted most heavily. That is a description of an aim rather than a ranking factor you can implement, but it does have a practical translation: content should contain something the writer could only know by having done the work.

Which is why our briefs are specific about sourcing and about who needs to be interviewed. The gap between adequate content and content worth linking to is almost always a concrete detail — a number from real experience, a failure mode, a decision that turned out badly — and no amount of research tooling generates that. It comes from your team, which is why a content programme that never books time with your practitioners plateaus at competent and stays there. Our content marketing practice runs that production side at volume.

The other half of the content programme is maintenance, and it is the half nobody sells. Pages decay as competitors publish, information ages and intent shifts. A mature programme spends a substantial share of its effort updating existing pages, consolidating overlapping ones, and occasionally removing pages that add nothing. Consolidation in particular is frequently the highest-return action available on an established site, and it involves publishing less rather than more, which is presumably why it is under-recommended.

Links, and what we will not do to get them

Links still matter, and the market for acquiring them is largely a market for liabilities.

Google’s spam policies name the practices explicitly: link spam, including buying and exchanging links at scale; scaled content abuse, meaning generating many pages of low value to manipulate rankings; site reputation abuse, where third-party content is hosted on a reputable domain primarily to exploit its ranking signals; expired domain abuse; cloaking; and doorway pages. We do none of them, and the reason is not only ethical. Links acquired at scale are increasingly likely to be discounted rather than penalised, which means the common outcome is not a dramatic penalty but a quiet waste of budget.

What works is slower. Publish something genuinely worth citing — original data, a useful tool, a clear position on something contested — and then do the unglamorous work of finding the people who write about your sector and giving them a reason to care. Fewer links, better sources, and a record of where every one came from. If you inherit a profile you cannot account for, that is worth auditing before it is worth adding to.

Structured data and entity signals

Structured data does not improve rankings. What it does is make a page eligible for rich results — the review stars, FAQ expansions, product details, breadcrumb trails and event listings that change how a result looks and therefore how often it is clicked. Google supports a specific and finite list of types, and markup outside that list is valid Schema.org and completely inert in Search.

So the useful work is narrow: identify which supported types your content genuinely qualifies for, implement them accurately, and validate them rather than assuming. Marking up content that is not visible on the page, or claiming a type the page does not actually represent, is a policy violation rather than a clever tactic. Where structured data does something quieter but valuable is in entity clarity — Organization markup, consistent naming, sameAs references — which helps disambiguate who you are across the web. For businesses with physical locations, that overlaps heavily with Google Business Profile work, covered on our local SEO page.

AI Overviews and answer engines

The most common question we now get is whether AI answers have changed the discipline. Google’s own position is unambiguous: there are no additional requirements to appear in AI Overviews or AI Mode. Be indexable, be eligible for snippets, follow the same policies, and publish helpful content.

That is genuinely true and slightly incomplete, because two things have changed in practice. The first is measurement: an answer assembled from several sources can satisfy a query without a click, so impressions can hold or rise while clicks fall, and reading either number alone will mislead you. The second is content shape. These features use a query fan-out technique — issuing multiple related searches across subtopics and data sources to assemble a response — which rewards covering a topic properly, with clear structure and quotable, self-contained statements, over publishing a thin page per keyword variant.

That is the whole of the honest version. Where a vendor sells you an AI search package, the reasonable question is which part of it is not simply the fundamentals with new packaging. We maintain a page on generative engine optimisation precisely so the measurement and structure implications can be discussed without pretending the ground rules were rewritten.

Where SEO splits into genuinely different work

Four areas need their own method rather than a variation on a general programme. Technical work deals with crawling, rendering, indexing and site health at a scale where tooling and log data matter more than judgement about wording. Local search is effectively a different index with different signals, dominated by proximity, the business profile and reviews, and a national content strategy will not move it. Ecommerce SEO is mostly a problem of catalogue architecture, faceted navigation and duplicate variants rather than of content quality. And answer engine work is largely about measurement and content structure. Each has its own page and its own method, and the pages are worth reading if you suspect one of them describes your actual situation.

What good measurement looks like

Search Console for what happens in the results, Google Analytics 4 for what happens afterwards, joined and segmented by page group and intent so you can see whether the pages that matter are gaining. Reporting monthly in writing, quarterly in depth, with the changes made and the changes proposed stated plainly.

The composition of that work changes as a programme matures, and it is worth knowing in advance so that a quiet reporting period does not read as a stalled one. Early on, most of the effort is diagnostic and structural: crawling, index coverage, the keyword and intent map, the redirect and architecture decisions. Very little of that produces a chart anyone enjoys looking at, and almost all of it determines whether the later work can succeed. The middle phase is production — briefs, pages, structured data, internal linking — where volume of output is the visible metric and quality is the one that matters. The mature phase is the least glamorous and often the most productive: refreshing pages that have decayed, consolidating overlapping ones, pruning what dilutes, and pursuing the links that a genuinely good asset can now earn.

A useful question to ask any agency, ours included, is what they expect to be doing once the obvious fixes are done. If the answer is more of the same content at the same rate forever, the programme has no second act.

And one commitment that costs us something: if the evidence says the programme is not working, or that your constraint is the offer rather than the traffic, we will tell you. Organic search is a long investment and the wrong time to discover it was misdirected is at the end. If you want a second opinion on a programme already running — yours or someone else’s — that is a conversation worth having before the next tranche of budget is committed. Organic search also rarely stands alone: it sits inside a wider acquisition picture where paid media, content and the quality of the landing experience all decide whether the traffic you earn is worth anything.