Every site I inherit arrives with the same request phrased slightly differently: “we need more traffic.” My first job is to slow that sentence down, because organic traffic is an outcome, not a lever. You cannot pull it directly. You can only fix the things that produce it: crawlability, relevance, authority, and the quality of the answer you give when someone lands.
What follows is the exact sequence I run, in the order I run it, and the reasoning behind that order. It is deliberately front-loaded with diagnosis. Roughly the first three weeks of any engagement produce no new published content at all, and that is the point. Publishing into a broken foundation is the most expensive mistake in SEO.
Step 0: Define What “Traffic” Actually Means for This Business
Before touching a single tool, I want two numbers from the client: what a conversion is worth, and what percentage of current organic sessions convert.
This changes everything downstream. A personal injury firm might need 400 sessions a month to hit its revenue target because a single signed case is worth five figures. An e-commerce brand with a $38 average order value needs volume. If I optimise the law firm for volume, I will fill the funnel with students writing essays and competitors doing research, then report a beautiful traffic graph and a flat revenue line.
So Step 0 produces a one-page definition:
- The commercial goal (leads, revenue, bookings, demo requests)
- The pages that produce it today
- The pages that should produce it but do not
- The traffic segments that will never produce it, which I will happily deprioritise
I also reset the definition of the KPI itself. In 2026, “sessions” alone is an incomplete metric. AI Overviews, AI Mode, and assistant-based search resolve a large share of informational queries without a click. A site can gain visibility and lose sessions in the same quarter. So I track a composite:
| Metric | What it tells me |
|---|---|
| Qualified organic sessions | Traffic from pages with commercial or transactional intent |
| Impressions and average position by query cluster | Whether visibility is growing even when clicks are not |
| Citation share in AI answers | Whether the brand is being used as a source, not just ranked |
| Assisted conversions from organic | The revenue contribution the last-click model hides |
| Branded search volume | The long-term compounding signal that a strategy is working |
If the client only wants to look at one chart, I make it qualified sessions plus branded search. Those two together are very hard to fake.
Step 1: Baseline Audit, Before Any Recommendation
I do not give advice in week one. I collect evidence.
Crawl the site properly. I run Screaming Frog with JavaScript rendering enabled and connect it to GSC, GA4, and Ahrefs APIs so the crawl output carries performance data alongside technical data. Then I run Sitebulb over the same site because its hint prioritisation surfaces architectural problems that a flat crawl list buries. Two crawlers, two perspectives, one merged issue register.
Pull 16 months of Search Console data. Not the 3-month default view. I need seasonality and I need the before-and-after of any algorithm update or site migration in the last year. I export at query and page level and build the analysis in a spreadsheet rather than the GSC interface, because the interface caps and samples in ways that hide the long tail where most of the opportunity lives.
Segment the loss. If traffic has declined, I want to know precisely what declined. Was it one template? One content cluster? Informational queries only, which would point to AI answer displacement rather than a penalty? A specific country? A specific device? “Traffic is down 30%” is not a diagnosis. “Traffic is down 30% because the blog’s top-of-funnel cluster lost 60% of clicks while impressions held steady” is a diagnosis, and it prescribes a completely different response.
Map the competitive set. Not the client’s list of business rivals, which is usually wrong. The SERP competitors: whoever actually occupies the top three positions for the money terms. I pull their content inventory in Ahrefs and SEMrush, and I note their site structure, because structure is often the real differentiator and it is the thing clients least expect.
Check what AI systems say about the brand. I query the major assistants directly for the client’s core commercial questions and record who gets cited. This is now a standard part of my audit. If a competitor is being quoted as the authority in generated answers, that is a visibility problem that no ranking report will ever show you.
Deliverable at the end of Step 1: a prioritised issue register scored by impact and effort, not a 90-page PDF that nobody reads. Every line has an owner and an estimated traffic impact.
Step 2: Fix the Technical Foundation
Technical SEO rarely creates growth on its own. It removes the ceiling that prevents growth. I sequence it first because every later step compounds through it.
Indexation before anything else. I compare the pages I want indexed against the pages actually indexed. The gap in both directions matters. Missing pages mean lost revenue. Junk pages that should not be indexed (filtered URLs, tag archives, internal search results, staging subdomains, thin location doorways) dilute crawl budget and, more importantly, dilute the site’s topical identity. I have seen sites gain 20% in organic traffic purely from removing 4,000 worthless indexed URLs.
Crawl efficiency. Broken internal links, long redirect chains, orphan pages, parameter sprawl, and pagination handled badly. On large sites I check the server logs. Log files tell you what Googlebot is actually doing, as opposed to what you assume it is doing, and they routinely reveal that half the crawl budget is being spent on URLs nobody wants ranked.
Core Web Vitals, but proportionally. Performance is a real ranking factor and a very real conversion factor, so it earns attention. It is not, however, the reason a site is invisible. If the site is at 2.8 seconds LCP and the competitor at 2.1, that is not why they outrank you. I fix the obvious wins (image compression, lazy loading below the fold, render-blocking scripts, caching) and then move on rather than chasing a perfect score.
Rendering. If key content or internal links only appear after client-side JavaScript execution, I test what is actually in the rendered DOM. Many AI crawlers render less thoroughly than Googlebot does, which means a JavaScript-dependent site can rank acceptably in traditional search and be functionally invisible to assistant-based retrieval.
Mobile parity. Not “does it look fine on a phone,” but does the mobile version contain the same content, links, and structured data as desktop. Any content hidden on mobile is content that may not be counted at all.
Step 3: Build the Keyword and Intent Map
Most keyword research fails because it produces a list. A list has no structure, so it gets published as a pile of disconnected posts that compete with each other and signal nothing coherent to a search engine.
What I build instead is a map: topic clusters organised around entities and the questions real buyers ask.
Start with the money terms. Every business has 10 to 30 queries where the searcher is ready to transact. These get dedicated, well-built pages. Nothing else takes priority over these.
Expand into the problem space. For each money term, I map the questions that precede it. Someone does not wake up searching “commercial hvac maintenance contract melbourne.” They start with a symptom, a cost question, or a comparison. Those upstream queries are where topical authority is built and where AI systems find the material they cite.
Cluster by intent, not by keyword similarity. “best crm for small business” and “crm pricing comparison” are lexically different and functionally identical. They belong on one page. Meanwhile “how to migrate crm data” looks related and serves a completely different person at a different moment, so it gets its own page. Splitting intent across pages is the single most common cause of cannibalisation I see.
Check difficulty honestly. A domain with 40 referring domains does not compete for a head term against a domain with 8,000, regardless of how good the content is. I pick a winnable perimeter first, build authority there, then expand. Early wins matter for retention, not just for reporting. Momentum funds the harder work.
Format the output cleanly. My keyword deliverables use lowercase, no hyphens, one row per target query, with columns for cluster, intent, monthly volume, current position, target URL, and gap type. It should be usable by a writer without further explanation.
Step 4: Fix Existing Content Before Producing New Content
This is where the fastest gains almost always live, and it is the step clients most want to skip.
Every mature site has three categories of underperforming content:
Striking distance pages. Positions 5 to 20, with meaningful impressions and poor click-through. These pages have already earned relevance signals. They usually need better intent alignment, a stronger opening answer, updated data, additional subtopics the SERP clearly rewards, and internal links from stronger pages. Turning a position 11 into a position 5 often doubles the clicks on that page, and it takes a day of work rather than a month.
Cannibalised clusters. Two or more pages competing for the same intent, splitting links and confusing relevance. I consolidate: pick the strongest URL, merge the useful content into it, redirect the rest. Consolidation almost always outperforms leaving both live.
Decayed content. Pages that ranked well and slid. Usually the query evolved, or the SERP changed format, or the content contains a date-stamped fact that is now wrong. A refresh cycle on decayed content is the highest ROI activity in most content programmes.
Only after this pass do I greenlight new production, and then I prioritise by the gap analysis: clusters where competitors rank and the client has nothing at all.
On quality, I hold one standard: the page must contain something that could not have been generated from the other nine results on that page. Original data, real client examples, a practitioner’s opinion, a photograph of actual work, a calculator, a documented process. Search engines and AI systems are both increasingly good at recognising synthesised restatement. Being genuinely additive is now the durable strategy, not a nice-to-have.
Step 5: On-Page Optimisation and Entity Clarity
With the map set and the content plan agreed, the on-page layer is mostly craft.
Title tags carry the primary query in natural language and a differentiator. I write them for the click, not for the crawler, then verify Google is not rewriting them, which is itself a signal that the title missed the query.
H1 and heading structure should read as a logical outline of the page. If someone read only the headings, they should understand the argument. This matters more than it used to, because heading structure is one of the ways passage-level retrieval systems decide which chunk of a page answers a question.
The opening answer. Every page opens with a direct, self-contained answer to the query it targets, in roughly 40 to 60 words, before any preamble. This serves the impatient reader, the featured snippet, and the AI system extracting a citation, all at once. It costs nothing and it is the highest-leverage formatting decision on the page.
Structured data, implemented properly. Organization and LocalBusiness for the entity, Article or BlogPosting for editorial, Product and Offer for commerce, FAQPage where genuinely applicable, Service for service pages, Person for author entities, and BreadcrumbList throughout. I validate every implementation in the Rich Results Test and the Schema Markup Validator, and I connect the entities with sameAs and @id references rather than leaving isolated blocks on each page. Connected schema describes a business. Disconnected schema describes a page.
Entity consistency. Name, address, phone, service descriptions, and founding details should be identical across the site, Google Business Profile, and major citation sources. Inconsistency here quietly undermines everything else, because it makes the entity harder to resolve with confidence.
Step 6: Optimise for Answer Engines, Not Only Search Engines
This is the layer most strategies still treat as optional, and it is the one that will separate sites over the next two years.
Traditional SEO asks: will this page rank. Answer engine optimisation asks: when a system generates an answer, will it use our content and name us as the source.
What actually works:
Write extractable passages. AI systems retrieve chunks, not documents. A 300-word paragraph containing four ideas is harder to cite than four clean 75-word passages each making one claim. I structure content so that any given section stands alone and remains accurate when quoted out of context.
Answer the question in the question’s own language. If people ask “how much does x cost in y,” the page should contain that phrasing as a heading and then answer it immediately. Coy content that dances around pricing does not get cited.
Be specific and verifiable. Numbers, dates, named processes, ranges, and conditions. Vague content is not citation material. “Costs vary depending on your needs” is a wasted paragraph. “Typically between $2,400 and $6,800, depending on site size and whether structural work is required” is citable.
Publish original information. Proprietary data, survey results, aggregated case outcomes, and documented methodology give AI systems a reason to cite you specifically rather than any of your competitors saying the same generic thing.
Establish author and organisation credibility explicitly. Real bylines, real credentials, linked profiles, and a clear About page that states who the organisation is, what it does, where it operates, and since when. Machines resolving trust need this stated plainly, not implied through design.
Get mentioned in the sources AI systems favour. Reddit threads, industry publications, review platforms, and comparison sites are disproportionately represented in generated answers. Being the recommended answer inside those sources is now a legitimate visibility channel.
I measure this by running a fixed set of prompts monthly across the major assistants and logging citation share against competitors. It is manual, it is imperfect, and it is far better than assuming.
Step 7: Internal Linking and Site Architecture
Internal linking is the most underused tool in SEO. It is free, entirely under the client’s control, and it works quickly.
The principles I apply:
- Every important page should be reachable within three clicks of the homepage
- Money pages should receive the most internal links, from the most authoritative pages
- Anchor text should be descriptive and varied, not “click here” and not the same exact phrase 200 times
- Cluster pages link up to the pillar and across to siblings, so the cluster reads as a coherent topic rather than scattered posts
- Orphan pages get adopted or removed
On larger sites I model this properly: export the internal link graph from the crawl, identify which URLs are accumulating internal authority, and compare that against which URLs should be. The mismatch is usually dramatic. A site’s most linked internal page is very often its privacy policy.
Step 8: Build Authority Off-Site
Links still matter. They matter less than they did in 2015 and considerably more than the “links are dead” crowd claims.
What I pursue, in order of preference:
Digital PR and original data. Publish something worth referencing. Survey your customers, aggregate your internal data into an industry benchmark, or take a defensible position on a live debate in your sector. This earns links from publications no outreach email will ever unlock.
Genuine industry relationships. Suppliers, partners, associations, sponsorships, local institutions, and trade bodies. Unglamorous, durable, and usually available in a week.
Expert contribution. Getting the client’s actual expert quoted in trade press. This builds the person as an entity, which increasingly matters as much as the domain.
Reclamation. Unlinked brand mentions, broken backlinks pointing at dead URLs, and lost links from redirect mishaps. Cheap and immediate.
What I avoid: bought link packages, private blog networks, and mass guest posting on sites that exist only to sell links. The risk profile is bad and the effect is temporary.
For local businesses I treat Google Business Profile as a primary asset rather than an afterthought: complete categories, service areas, products, real photographs, a consistent posting cadence, and an active review generation process with owner responses. For local intent queries, GBP frequently outperforms the website itself.
Step 9: Measure, Report, and Iterate
I set up reporting before the work starts, not after the client asks for it.
The stack: GA4 configured with proper conversion events, Search Console connected at the domain property level, rank tracking segmented by cluster, and a Looker Studio dashboard the client can read without me in the room. Annotations on every significant change, so that three months later we can attribute movement to a specific action instead of guessing.
The reporting cadence I use:
- Weekly (internal): anomaly checks. Indexation drops, crawl errors, ranking volatility, traffic cliffs.
- Monthly (client): what was shipped, what moved, what is next. Traffic and conversions against baseline, cluster-level ranking movement, and citation share in AI answers.
- Quarterly: strategic review. Is the thesis still right? Which clusters are outperforming and deserve more investment? Which are not and should be cut?
The discipline that matters most: kill what is not working. SEO programmes fail more often from stubborn commitment to a plan than from a bad plan. If a cluster shows no traction after two quarters of good execution, the market is telling you something.
A Realistic 90-Day Rollout
| Timeframe | Focus | Expected outcome |
|---|---|---|
| Weeks 1 to 3 | Audit, analytics setup, keyword map, prioritised issue register | No traffic change. Full clarity on what is broken and what is winnable. |
| Weeks 3 to 6 | Technical fixes, indexation cleanup, internal linking pass, schema implementation | Improved crawl efficiency and early movement on striking distance pages. |
| Weeks 5 to 10 | Content refresh programme, cannibalisation consolidation, on-page optimisation of money pages | First meaningful traffic and conversion lift, typically 10% to 30% on refreshed pages. |
| Weeks 8 to 12 | New content production against gap analysis, AEO formatting standards applied, digital PR campaign launched | Cluster expansion, first citations in AI answers, initial link acquisition. |
| Month 4 onward | Compounding production, authority building, quarterly strategy review | Sustained growth curve. Most sites see the inflection here, not before. |
Anyone promising a transformation in 30 days is either sitting on unusually broken technical foundations, which does happen, or they are not being honest with you.
The Five Mistakes That Cost the Most
- Publishing before diagnosing. New content on a broken foundation buries the problem instead of solving it.
- Chasing volume over intent. Traffic that cannot convert is a cost, not an asset.
- Treating each page as an island. Search engines and AI systems evaluate topical coherence across a site, not page by page.
- Ignoring what already exists. The fastest wins are almost always in pages that already rank on page two.
- Optimising only for the blue links. A meaningful share of search demand now resolves inside generated answers. If your content is not structured to be extracted and cited, that demand goes to whoever structured theirs.
The Underlying Thesis
Strip away the tactics and the strategy reduces to four questions:
Can they find it? Can they understand it? Do they trust it? Is it the best answer available?
Technical SEO answers the first. Structure, schema, and clarity answer the second. Authority, credibility, and consistency answer the third. Genuinely useful, original content answers the fourth.
Everything in this playbook is in service of those four questions. The channel keeps changing, the interfaces keep changing, and the questions do not. That is what makes this approach durable while individual tactics come and go.
