The complete SEO guide 2026

Everything starts here. It does not matter how well built the site is technically if it answers questions nobody asks — or answers the wrong way on questions people actually ask.

Understand the intent behind the search

Every search has an intent, and Google has become very good at matching it. Four basic types cover most of it:

Informational

“what does a website cost” – the person wants to learn something, not buy. Yet.

Commercial

“best web agency london” – the person is comparing options ahead of a decision.

Transactional

“web agency fixed price quote” – the person is ready to act now.

Navigational

“brucy contact” – the person is looking for a specific page belonging to someone they already know of.

Search your most important keywords and look at what actually ranks. If the top ten is full of guides, both Google and the user expect a guide — not a product page. Matching the wrong intent is the most common reason good content never ranks.

Do the research properly

Start from the business, not from the tool. List the problems you solve, the words customers use themselves — they almost always differ from the industry jargon — and the questions you get in sales meetings. Every recurring question is a keyword.

Then add data. Google Search Console shows what you already appear for, Google's autocomplete and “related searches” show how people phrase things, and tools like Ahrefs, Semrush or Keyword Planner give volumes and competition.

Then prioritise. A keyword with fifty searches a month and crystal-clear buying intent is usually worth more than one with five thousand searches and vague intent. Build a keyword map: one primary keyword plus a handful of variants per page, and never two pages competing for the same keyword.

Technical SEO is invisible when it works and devastating when it does not. A site Google cannot crawl, render and index correctly will never rank — however good the content is.

Can the search engines reach your content?

robots.txt

Belongs in the domain root, UTF-8 encoded, and should point to your sitemap. It must never block CSS or JavaScript – Google has to be able to render the page as a user sees it. A classic disaster is a Disallow: / left over from the development environment, silencing the whole site. Check it before anything else.

XML sitemap

Should contain all indexable pages and only those. No pages that return 404, redirect, or carry noindex. Submit it in Search Console and keep it updated automatically – modern frameworks generate it for you.

Status codes

200 for working pages, 301 for permanent moves, 404 or 410 for removed content, 503 during temporary maintenance. Avoid redirect chains: one hop at most. Every extra hop leaks link equity and crawl budget.

Crawl budget

Rarely a problem below a few thousand pages, but filter pages, session parameters and infinite URL combinations can swallow it even on smaller sites. Clean up with canonical, noindex or robots rules.

Do the right pages end up in the index?

Canonical tags

Absolute URLs, self-referencing on the original page. They must never point at a page that redirects, returns 404 or carries noindex – that sends contradictory signals.

noindex

On pages that add no search traffic: thank-you pages, internal search results, filtered lists, test pages.

Soft 404

Pages that say “not found” but return 200 confuse Google and waste crawl budget. Removed pages should return a real 404 or 410.

Duplicate content

Consolidated with canonical or 301. Common culprits: http versus https, www versus non-www, duplicates with and without a trailing slash, print versions.

Orphan pages

Pages with no internal links to them are rarely found and almost never rank. Every important page should be reachable through the site's own structure.

JavaScript and rendering

If the site is built in React, Vue or something similar, the rendering strategy is an SEO decision. Serve the content server-rendered or statically generated so that it is in the HTML response directly.

Pure client rendering means Google has to run your JavaScript to see the content. That works sometimes, slowly – and many AI crawlers run no JavaScript at all. Content that only exists after JS execution is invisible to large parts of the AI search landscape.

Verify with URL Inspection in Search Console: does the rendered HTML look like what users see? We build everything in Next.js with server rendering for exactly this reason – the content is in the HTML from the first byte.

HTTPS and security

HTTPS across the whole site, a valid certificate, no mixed-content warnings. It is table stakes in 2026 – but still worth a line on the checklist, because certificates expire and old image links creep back in.

Google measures the user experience with three field metrics, collected from real Chrome users. They are a ranking factor, but more importantly than that: slow sites lose visitors before the page has even loaded.

The three metrics

LCP = Largest Contentful Paint, INP = Interaction to Next Paint, CLS = Cumulative Layout Shift.

The thresholds have to be met by the 75th percentile of your visitors. So it is not enough that the site is fast on the office wifi. Measure field data in Search Console and PageSpeed Insights, not only lab data, and break it down by device type – mobile is almost always worse.

How to meet the thresholds

LCP

Optimise the largest element, usually the hero image or the headline. Serve images in WebP or AVIF, at the right size for the screen, and set fetchpriority="high" on the LCP image – never lazy-load it. Fast server response through good hosting and a CDN.

INP

Minimise JavaScript on the main thread. Defer non-critical scripts, break up long tasks and be strict with third-party scripts. Chat widgets, tracking pixels and A/B testing tools are notorious INP killers. Every script has to earn its place.

CLS

Always set width and height, or aspect-ratio, on images and embeds. Reserve space for ads and dynamic content. Load fonts with font-display: swap plus preload so the text does not jump.

On-page is the craft: making sure every single page is built so that both search engines and people immediately understand what it is about and why it is worth clicking.

The title tag

Still one of the strongest on-page signals, and the first – often only – thing people read in the search result. Primary keyword early, unique per page, roughly 50–60 characters so it is not truncated, and written to be clicked. “Web agency with a fixed monthly price – site, hosting and support” beats “Home” every day of the week.

Meta description

Not a direct ranking factor, but it drives click-through – and a page that gets clicked more gets more chances. Roughly 150–160 characters, summarise the value, end with a reason to click. Google rewrites it sometimes, but a good original usually wins.

Heading structure

One H1 per page, containing the primary keyword, describing the page's content. H2 and H3 in a logical hierarchy – do not jump from H1 to H4. Write headings as questions where it fits; that matches both how people search and how AI engines pick answers.

URLs

Short, descriptive, in the site's own language, with hyphens between words and without dates, parameters or ID numbers. And once a URL ranks: leave it alone. Every change requires a 301 and risks a loss.

Internal linking

The most underrated factor. Internal links show Google which pages matter most and how topics connect. Link from content-rich pages to your money pages, use descriptive anchor text – “read our guide to technical SEO”, never “click here” – and build pillar pages with supporting articles that link up to the pillar and to each other.

Images

Descriptive file names rather than IMG_4711, and alt text that describes the image for someone who cannot see it. Natural language, no keyword stuffing. Alt text is also one of the few signals image search and language models have about what the image depicts.

Google's quality framework E-E-A-T – experience, expertise, authoritativeness and trustworthiness – is the yardstick. Does the content show real experience, genuine subject knowledge, authority in the topic, and an author you can trust?

Write the best answer that exists

For every page you want to rank with, ask the question: is this the best answer available for this search? If not – what is missing?

It is rarely about length. It is about answering completely, covering the follow-up questions, and adding something that is not already in the top ten: your own data, your own examples, your own images, a clearer explanation.

In an age when anyone can generate generic text in seconds, the unique is the only thing that stands out. Your actual client cases, your numbers, your experience from real projects. Filler without human substance ranks worse and worse – and is never cited as a source by an AI engine.

Show who stands behind it

A named author

On articles, with a real author page: role, background, experience.

A proper about page

Who you are, what you have done and why you know the subject.

Contact details and company registration

Anonymous sites are untrustworthy sites, to readers and search engines alike.

Source references

Whenever you use statistics or claims from elsewhere.

Own your topic

Google rewards sites that cover a topic in depth over sites with scattered individual articles. Build clusters: a pillar page for the broad topic, supporting articles for each sub-topic, all tightly linked.

Ten articles about your core subject beat a hundred articles about a bit of everything.

Keep the content alive

Date articles honestly and update the most important pages every three to six months with fresh numbers and years.

Clean up too. Pages that never got traffic and serve no purpose should be improved, merged or removed with a 301. A smaller and sharper site beats a large and diluted one.

Structured data is schema.org markup in JSON-LD that describes the content machine-readably: this is a company, this is an article, this is a question with an answer.

It can produce rich results – stars, FAQ expanders, breadcrumbs – and it helps both Google and AI models understand exactly what the page contains and who stands behind it.

The types most company sites need

Organization

Across the whole site: name, logo, contact details and sameAs links to your profiles. sameAs ties your entity together across the web – important for both the knowledge panel and the models' picture of who you are.

LocalBusiness

If you have a physical location or a local catchment area: address, opening hours, geo-coordinates.

Article or BlogPosting

On articles: author plus publication and update dates.

FAQPage

On pages with questions and answers. It also matches exactly how AI engines extract answers.

BreadcrumbList

For breadcrumbs in the search result.

Product, Offer and Review

For ecommerce: price, stock status and rating – producing stars and price information directly in the search result.

Service, VideoObject and Event

For service pages, video and events respectively.

Do it right

Use JSON-LD in a script block of type application/ld+json – Google's recommended format.

Only mark up things that actually appear on the page. Schema for content that is not there is misleading and can be penalised. Validate with the Rich Results Test and watch Search Console's enhancement reports for errors.

A growing share of all searches happens in AI interfaces: Google's AI Overviews and AI Mode, ChatGPT, Perplexity, Copilot. There are no ten blue links there – there is one answer, with a handful of cited sources.

Becoming one of those sources is called AEO or GEO. It is not a separate discipline but an extension of good SEO, though there are specific things to do.

Write so that an AI can cite you

The answer first

Answer the page's main question in the first few sentences – short, factual, complete. Then expand. A page that builds up to the answer in paragraph fourteen does not get cited.

Question-shaped headings

An H2 like “What does a website cost in 2026?” matches exactly how users ask, and turns every section into a self-contained, citable block.

Concrete numbers with a source

Attributed statistics get cited more often than vague claims. Your own first-party data is gold: it exists nowhere else, so anyone who wants to repeat it has to point at you.

Self-contained paragraphs

Every paragraph should make sense lifted out of its context. Tables, bullet lists and definitions are easy for language models to extract correctly.

Technical access for AI crawlers

The AI services crawl with bots of their own: GPTBot and OAI-SearchBot from OpenAI, ClaudeBot from Anthropic, PerplexityBot, and Google-Extended for Google's AI training. If you want to appear in AI answers they have to be allowed in robots.txt – only block them if you deliberately want to stay out.

Note the difference between training bots and search bots. You can, for instance, block GPTBot, which is training, while allowing OAI-SearchBot, which is search and citation, if you want to appear in ChatGPT's search mode without contributing to model training.

At least as important: many AI crawlers do not render JavaScript. If the content is client-rendered it is invisible to them. Another reason for server rendering. The llms.txt file has been proposed as a standard, but Google have been clear that they do not use it – put the effort into real HTML and schema instead.

Build entity and mentions

AI models recommend brands they know of: the ones that appear in directories, on review sites, in trade media and in articles. Digital PR – being mentioned, not only linked – carries more and more weight.

Consistent entity information, meaning the same name, description and details everywhere, plus Organization schema with sameAs, makes you identifiable as an entity rather than as a random site.

Test yourselves

Ask ChatGPT, Gemini and Perplexity what your customers would ask. “Best web agency with a fixed monthly price?” “What does an ecommerce site cost?” Do you appear? Are the competitors cited?

Make it a monthly routine. It is your new ranking report.

If you sell to a region or a city, local SEO is often what gives the fastest effect. “web agency gothenburg”, “plumber near me”, “accountant in my area” – a lot is decided here in the map pack above the ordinary results.

The Google Business Profile is the hub

Claim the profile and fill it in to a hundred per cent: the right category – the single most important choice – services, description, opening hours, images and posts.

Reply to every review, including the negative ones. Especially the negative ones. Ask satisfied customers for reviews systematically; volume, rating and recency are heavy local ranking factors.

NAP consistency and local pages

Name, address and phone number should be identical everywhere: on the site, at Google, in local business directories, in company registers and in trade directories.

If you operate in several cities: build a real page per city with unique content – local cases, local references, local knowledge. Ten copied city pages with only the place name swapped out do not work, and can do harm.

Links from other sites are still one of the strongest ranking signals – a vote that your content is worth trusting. But a link from a respected industry site is worth more than a hundred directory links.

How to build links that mean something

Link-worthy content first

Nobody links to your service page voluntarily. People link to data, studies, guides, tools and unique insight. A thorough guide is exactly that kind of asset.

Digital PR

Your own data plus an angle journalists care about produces mentions in the trade press. That gives both links and the brand mentions AI models weigh in.

Natural sources

Industry bodies, supplier and partner pages, local business sites, and guest articles where you actually have something to say.

Keep it clean

Bought link packages, link farms and automated junk links are wasted money at best and a manual action at worst. Build nothing you would not dare show Google.

SEO is not a project with an end date, it is a process. Without measurement you do not know what works, and without a routine even the best optimised site decays.

The toolbox

Google Search Console

The most important tool, and free: clicks, impressions, positions, indexing status, Core Web Vitals and links. If you do not live here regularly you are flying blind.

Google Analytics 4 or a privacy-friendly alternative

Plausible and Matomo do the same job. The question they answer: what do visitors do after the click, and do they convert?

PageSpeed Insights and Lighthouse

Performance analysis per page, with both lab and field data.

Screaming Frog or Sitebulb

Crawls the site the way Google does and finds technical errors in bulk.

Ahrefs or Semrush

Positions over time, competitor analysis and link profiles.

The routine that keeps the site in shape

Every week

Search Console: indexing errors, coverage warnings and drops in clicks.

Every month

Positions against the target keywords, Core Web Vitals, the AI visibility test from section 7, and publishing or updating according to the content plan.

Every quarter

A full technical crawl, a content audit – what should be updated, merged or removed? – a link profile review and competitor analysis.

When traffic suddenly drops

Check Search Console first: is there a manual action or an indexing error? Then check whether a core update has rolled out. After that, look at technical changes on the site – a deploy that happened to add noindex, or pages that were removed.

Change nothing in a panic before you know the cause. That is how small problems become big ones.

After many walkthroughs of company websites it is the same things that keep coming back. Check that you are not doing any of these:

1. noindex or Disallow left over from the development environment

   The site is invisible and nobody has noticed. This is the cause more often than you would think.

2. The homepage is supposed to rank for everything

   And therefore ranks for nothing. One page, one primary keyword.

3. Titles like “Home” and “About us”

   Wasted positions on the most important lines of text on the site.

4. Client-rendered content

   Pretty in the browser, empty for crawlers and AI engines.

5. The site built and then “finished”

   No measurement, no updates, slow decay.

6. Content written for search engines rather than people

   Keyword-stuffed copy nobody wants to read ranks worse than ever.

7. All eggs in the Google basket

   No thought for AI search, where the customers are already asking their questions.

Keywords and intent: know what you should rank for and why. Technical SEO: crawlable, indexable, server-rendered, https. Performance: LCP under 2.5 seconds, INP under 200 milliseconds, CLS under 0.1. On-page: unique titles, one H1, internal linking, alt text.

Content and E-E-A-T: the best answer, a real author, topic clusters, an update routine. Structured data and AI search: schema, the answer first, AI crawlers allowed, test in ChatGPT. Local, links and measurement: the Business Profile, digital PR, Search Console as a habit.

It is a lot. It is supposed to be – that is why most people do not do it, and why it works for those who do.