If Your SEO and AEO Strategy Starts With Keywords, It Starts Too Late

Filed under Search & AI Discovery

It is surprisingly easy to make an SEO strategy look finished.

Open a keyword tool. Export a few thousand rows. Add intent labels, clusters and some tasteful conditional formatting. Put the largest numbers near the top.

There. Strategy.

Except it is not.

It is a list of observed searches with excellent table manners.

I like keyword data. I use it all the time. The problem starts when we ask it to make decisions it cannot make: which market matters to the business, which customer situation we can genuinely help with, which questions influence a buying decision, and what we can answer with enough evidence to be useful.

The same problem has now arrived in AEO wearing a newer jacket. Replace keywords with prompts, rankings with citations, add a dashboard showing “AI visibility”, and the work can still begin too late.

For me, a useful SEO and AEO strategy connects the commercial problem, the intended audience, their real questions, the answers and evidence on the website, the systems that keep those answers current, and the signals that tell us what deserves attention next.

The individual parts are not the strategy. The connections are.

Strategy starts before anyone opens the keyword tool

Before deciding what to rank for, decide where better discovery needs to create value.

That sounds obvious. It is also the step most likely to be compressed into something like “target: enterprise” and then quietly abandoned.

I would start with one sentence:

We want to help [specific audience] understand and evaluate [specific problem or offer] when [buying situation], so they can [useful next decision or action].

For an illustrative software company, it could be:

We want to help digital and content leaders at multi-market organisations evaluate whether a new publishing platform can improve governance without creating an unrealistic migration project.

This is not award-winning positioning. Nobody is getting it tattooed on their forearm. It does something more useful: it gives the research a job.

Now a query about content governance, migration effort or local editor autonomy has a commercial context. A high-volume query about building a free personal blog probably does not.

Without that boundary, keyword research can produce a beautifully organised map of places the business has no reason to visit.

The strategy I am describing is a loop:

Commercial problem → ICP and buying situation → stakeholder decisions → buyer questions → answers and evidence → discoverability → measurement and learning

Search data can send you back to the questions. A sales objection can expose a missing answer. A content audit can reveal that the answer exists but is buried in a PDF last updated when everyone still thought the metaverse was inevitable.

The loop is less tidy than a funnel. That is unfortunate for PowerPoint and fairly normal for humans.

Your ICP needs to do more than fit inside a CRM field

An ICP made from industry, geography and employee count can help filter accounts. It gives a writer very little to work with.

“European software company with 500+ employees” may fit neatly in a database, but it has yet to ask a useful question.

For content strategy, I care about the situation around the account:

  • What changed and made the problem important now?
  • What are they doing today?
  • Where is the current setup failing or becoming expensive?
  • Who is involved in the decision?
  • What constraints could change the right answer?
  • What would make this organisation a poor fit?

Two companies can look identical in an ICP spreadsheet while needing completely different content. One may have an active migration project, internal developers and executive sponsorship. The other may have no owner, no budget and no appetite for changing anything until 2031.

They should not receive the same answer simply because both employ 700 people and have an office in Germany.

The buying situation produces better questions than the company description alone:

  • How much internal capacity will a migration require?
  • Can local teams keep autonomy without losing brand governance?
  • What must be migrated, archived or rewritten?
  • Which integrations need to work before launch?
  • Who owns the project after the exciting workshop ends?

That final question has ended more than one tidy strategy diagram.

ICP should not be sprinkled into every article like a mandatory seasoning. Its job is to help you decide which questions deserve a strong answer, which conditions the answer must cover, and which traffic is unlikely to matter.

A query is a doorway, not the full brief

Keyword research is good at showing how people enter an information space. It is weaker at showing everything they need once they get inside.

That is why I would never use one research source alone.

Search Console can show the queries and pages where your site already receives visibility. Sales calls can reveal buying triggers and objections. Support conversations expose the parts that become painful after purchase. Reviews contain the customer’s vocabulary, along with the occasional review written during what appears to have been a very difficult Tuesday.

Conversational data adds another useful layer: what people clarify after receiving an initial answer.

I analysed a sample of public multi-turn AI conversations to investigate what users add after the first prompt. In that exploratory sample, an estimated 62.6% of the usable follow-ups added context or a constraint that was not explicit in the first prompt.

The first prompt often named the subject. The follow-up revealed the environment, audience, limitation or level of specificity required to make the answer useful.

That does not make keyword research obsolete. It means “intent: informational” is not a personality profile.

I use Search Console as one input to this work. My workflow for turning Search Console queries into an AEO backlog connects the observed query to the surfaced page, the answer currently available, the commercial boundary and the evidence still missing.

The useful question is not simply:

What are people searching for?

It is:

Which relevant decision sits behind this question, and what would a credible answer need to include?

That extra sentence is where the spreadsheet starts becoming strategy.

Please do not turn every buyer question into an article

Once a team has collected 200 buyer questions, somebody will suggest 200 articles.

This is understandable. It is also how a useful research project becomes a content factory with nicer labels.

Some questions deserve a new article. Others belong on a product page, implementation guide, comparison page, case study or technical document. Several may be different phrasings of the same concern. Some reveal that the company needs a clearer internal answer before marketing should publish anything at all.

For each cluster, I would ask:

  1. Does this matter to an organisation we can realistically help?
  2. Does the answer reduce uncertainty around a meaningful decision?
  3. Is an existing page already being surfaced?
  4. Is the current answer absent, partial, vague or unsupported?
  5. Do we have the evidence to answer honestly?
  6. Where would the answer be most useful?

The output is an answer backlog, not an article quota.

Sometimes the right move is a new guide. Sometimes it is adding three honest paragraphs to a product page. Sometimes it is deciding that a query with impressive volume has almost nothing to do with the business and allowing it to live a happy life elsewhere on the internet.

The format comes after the answer.

A useful answer needs evidence and structure

Content can fail in two directions.

It can answer an important question badly. Or it can contain a good answer in a form that is difficult to find, extract or understand.

First, make the answer worth reading

“Easy to implement” is a good example of the first problem. It sounds reassuring until you try to use it for a decision. Easy for whom? Under which conditions? Who does the work? What tends to go wrong?

A useful implementation answer explains the stages, dependencies, responsibilities, common delays and conditions that change the estimate. If the company cannot support those claims, the next task is evidence collection, not more confident adjectives.

Then make the answer easier to retrieve

Once the answer is worth publishing, structure matters.

Important sections should still make sense when someone lands directly on them, reads them out of order or encounters a fragment in an AI-generated answer. That means descriptive headings, explicit subjects, consistent terminology, clear conditions and evidence placed close to the claim.

This is the thinking behind my AI Retrieval Content Checklist and the more practical AEO content audit.

But retrievability cannot rescue an irrelevant answer. Schema markup cannot perform CPR on a page that says nothing useful.

The commercial question and the structural work need each other. Otherwise, AEO becomes a formatting exercise applied with tremendous discipline to the wrong content.

The strategy has to survive contact with the CMS

This is the less glamorous part, which usually means it is where the real problem lives.

A team maps the buyer questions, agrees on the terminology and identifies the evidence. Then ordinary work resumes.

Pages become outdated. A new case study is published but never connected to the claims it supports. The best answer remains trapped in a webinar transcript. Three teams describe the same feature in four different ways. Nobody knows whether the implementation page was reviewed before or after the product changed.

Six months later, the strategy is still excellent. It is also mainly decorative.

I built a workflow that syncs Umbraco content into an Airtable library because I wanted the website content to be queryable alongside sources such as reviews, sales transcripts, support conversations and webinar material.

The specific stack is not the point. Use Airtable, another database or something held together by spreadsheets and unreasonable optimism. The operating pattern matters:

  • keep an inventory of important content;
  • preserve its URL, ownership and update context;
  • connect claims with available evidence;
  • make gaps and inconsistencies reviewable; and
  • keep human approval between analysis and publication.

A strategy that cannot survive publishing, maintenance and team handoffs is still a slide deck. It may be a very attractive slide deck. This does not improve the situation.

A dashboard can show movement. It cannot make the decision

Measurement is where everyone would like the loop to become clean and causal.

Search Console shows query and page movement. Analytics shows on-site behaviour under its own measurement model. CRM and sales feedback add commercial context. None provides a perfect account of how one answer influenced one buyer across search, AI tools, direct visits, forwarded links and the internal meeting where somebody pasted your article into a slide.

That does not mean we give up and measure vibes.

It means we record what we know:

  • why the page or question matters;
  • the current query and page evidence;
  • the weakness we intend to fix;
  • the change and publication date;
  • the expected reader or workflow outcome; and
  • when we will review it.

Then we look again and decide whether the evidence supports another change, a deeper investigation or no action.

I built Site Signal for this part of the work. It combines bounded Search Console evidence with separate GA4 or Matomo context and local page content. The result is an investigation queue, not a machine confidently ordering rewrites because a line went down.

A metric changing is an observation. A page rewrite is a decision. There should be some thinking between the two, even if the dashboard looks very serious.

What this looks like when the pieces connect

Return to the illustrative publishing-platform company.

The original brief

The original SEO brief might have been “rank for enterprise CMS keywords”. That would produce topics, volumes and competitors. Useful inputs, but still no clear decision.

The decision hiding behind the keywords

The connected version starts with the buying situation: a multi-market organisation has fragmented publishing workflows and is actively considering a platform change. Content leadership wants better governance. IT wants manageable architecture. The project owner fears a migration swamp. The executive sponsor needs a reason to prioritise any of it.

Search Console surfaces questions about implementation and governance. Sales calls reveal uncertainty about internal ownership. Delivery teams know which dependencies cause projects to stall. The current product page promises flexibility, because product pages enjoy promising flexibility, but says little about responsibilities or trade-offs.

The content action

Now the action is clearer.

Improve the existing implementation page with stages, dependencies, role clarity and links to relevant customer evidence. Connect it from the product and migration content. Keep the central explanation in accessible HTML rather than hiding it in a PDF. Record the related query-page baseline and the sales concern the change is meant to address.

Only create a separate article if the broader question needs room to be answered properly.

That is the connection I care about:

  • The ICP and buying situation decide which problem matters.
  • Stakeholder decisions reveal the useful questions.
  • Research tests whether those questions appear in the market.
  • Available evidence determines what can be claimed.
  • Content structure makes the answer easier to use and retrieve.
  • Measurement helps choose the next investigation.

Remove any one of those links and the work becomes less useful. Remove enough of them and you are back to a spreadsheet with lovely colours.

Start with one buying situation

You do not need to map the entire website, build a knowledge graph and convene a steering committee before trying this.

Pick one commercially important buying situation.

Identify the people involved and the decisions they need to make. Compare the questions visible in search data with what sales, support and customers hear. Inspect the pages already being surfaced. Find one answer that is absent, incomplete or unsupported. Improve it with the evidence you can genuinely provide. Record what changed and when you will review it.

Start with three question clusters, not 3,000 keywords.

The point is to make one part of the system work end to end. Once you can see the handoffs, you can expand without losing the reason the work exists.

SEO and AEO still require technical foundations, content craft and measurement. But those disciplines do not choose the market, understand the buying situation or manufacture evidence for you.

That is why a useful strategy starts earlier.

If your team has content, data and tools but no clear line between ICP, buyer questions, credible answers and the next commercial action, that is one of the problems I help untangle. You can see the ways I work on the Work with me page.

More writing connected by category or shared topics.