From ten blue links to quoted answers: how SEO changed, and what we do about it

Every few years someone declares SEO dead. The name on the invoice changes (AEO, GEO, AI SEO) and the slides get a new gradient. The underlying job does not. A firm still needs to be findable when a buyer looks, quotable when an assistant summarises the field, and decidable when a sceptical partner lands on the page.

What changed is the unit that gets rewarded. For a long time the prize was a position among ten blue links. Now a large share of demand ends without a click at all, and a growing share of shortlisting happens inside AI assistants and Google’s own generative answers.

Those systems do not “rank your brand” in the way a league table feels. They lift sentences. If your important claims never appear as ordinary text in the first HTML response, you are invisible to that layer of demand even when your Google rankings look respectable.

This piece is the longer, technical account of how we got here, what we still treat as best practice onsite and offsite, and how we use AI in the writing pipeline without pretending a model can own professional services proof. The shorter partner-facing cut of the same territory sits elsewhere in the slate. The buying-guide question (whether GEO or AEO should be a separate retainer) is a different article. Here we stay with method.

Has search been replaced by AI, or renamed?

Neither. Search widened. Google’s own public framing treats generative answers as part of the search experience, not as a rival product you buy beside SEO. AI assistants are a parallel discovery surface with different fetch behaviour: they take HTML, they do not run your JavaScript, and they prefer passages that stand alone. Often they also send a sceptical human to verify, which is why the destination page still has to convert.

So is “AI SEO” real?

As a label for “make the firm easy to quote”, yes. As a magic stack of prompts and schema that replaces crawlable pages, no. Vendors selling a bolted-on AI visibility product that never changes your service pages are selling theatre. The work that survives audit is still writing, information architecture, technical hygiene and off-site corroboration.

Do rankings still matter?

Yes, for the demand that still clicks, and as one input into how systems decide you exist. They are not the whole scoreboard. A firm can sit on page one for a practice term and still never be named when a buyer asks an assistant who to instruct. Quotability and convertibility sit beside rankings, not underneath them as an afterthought.

Hold the three tests in mind as you read on: findable, quotable, decidable. Rankings alone only speak to the first.

How SEO actually evolved

Early web discovery ran on directories, manners and good luck: reciprocal links, and pages that stated what they were. PageRank and the link graph professionalised that into a market. Firms learned that being mentioned by trusted sites was not vanity. It was evidence.

Then volume arrived. Content farms, doorway pages and spun copy treated keywords as a density problem. Google’s quality updates (remembered in the industry as Panda, Penguin and their successors) punished thin repetition and manipulative link graphs. The lesson that stuck for serious firms was dull and durable: write for a reader who might instruct you, on a URL that stays put, with links that would not embarrass you in front of a regulator.

Mobile-first indexing and HTTPS became baselines rather than differentiators. Understanding systems (BERT, MUM and related work) reduced the reward for awkward keyword stuffing and increased the reward for pages that answered the question the searcher meant. Featured snippets and “people also ask” panels already trained the market on passage-level retrieval years before AI assistants became a boardroom word. Zero-click behaviour then removed a large share of visits from analytics dashboards even when the brand was present in the answer box.

Why professional services felt each wave differently

Professional services firms felt each wave differently from ecommerce. A disputes practice never needed a thousand SKU pages. It needed a small set of practice areas that told the truth about who the firm acts for, and a reputation graph that matched how buyers already shortlist (directories, peer recommendation, later assistants). That is why “publish more blog posts” was always a weak substitute for fixing the page that receives the enquiry. The firms that treated SEO as brochure production kept paying for traffic into pages that could not close.

What survived every shift

AI Overviews and assistant citation are the latest turn of that same screw. They do not abolish the need for a coherent site. They punish the habits that already failed humans: hedging until nothing is citeable, locking the useful paragraph behind a tab, publishing a blog calendar instead of fixing the practice page, and measuring only what JavaScript can see. For the measurement half of that problem, see Your analytics cannot see AI assistants.

What survived every shift is boring on purpose: crawlable HTML, stable addresses, clear topical focus, mentions that corroborate who you are, and pages that answer a real question in language a partner would put their name on. What stopped working keeps returning under new branding. Doorway pages, thin spun copy, keyword density theatre, claims that never appear until a click, and content programmes that never touch the commercial spine.

What we implement onsite

The commercial spine comes first. For a professional services firm that usually means practice and service pages, contact and enquiry paths, people pages where reputation lives, and a small set of answers that buyers already ask. Insights deepen the argument. They do not replace the spine. One canonical URL per distinct answer beats five near-duplicates fighting each other.

Claims have to sit in the first HTML response. Who you act for, what you will not take on, what happens after contact, and how work starts are operational facts, not marketing garnish.

Assistants that fetch without running scripts will quote those sentences or they will quote somebody else. Humans experience the same failure more quietly: they read, they cannot tell, they leave.

We treat Website Development as the ceiling when the site cannot expose those claims without a rebuild, and we say so early rather than selling a visibility retainer into a wall. The readiness check that looks at quotability before spend scales is AI Search Readiness.

Does technical hygiene or schema decide it?

Technical hygiene is hygiene. Indexability, honest robots and sitemaps, internal links that pass meaning rather than only PageRank folklore, and Core Web Vitals as a floor, not a strategy. Structured data earns its place when it matches visible content. It does not invent a claim the page refuses to print. We will not recommend schema theatre as a substitute for a sentence a consultant can defend.

Why named people and conversion count as SEO

People pages and author pages matter more than many firms admit. Assistants and buyers both use named humans as a trust shortcut. An unsigned service page reads as marketing. A named author with a stable URL, consistent job title and links from the pages they stand behind reads as a firm that will answer the phone. That is editorial and technical work at once: the bio has to exist in HTML, and the claims on the commercial pages have to match what that person would say in a meeting.

Conversion is part of SEO in this market. An assistant that sends a sceptical visitor to a brochure that never asks for the instruction has done you no favour. Demand pages get rewritten before paid programmes scale into them. That sits in the Conversion stage for a reason, and the writing work itself is Content and Editorial. Visibility without a decidable destination is expensive proof of a writing problem.

What we implement offsite

Offsite work is corroboration, not decoration. The firm’s name, people and service language should match across directories, Google Business Profile, LinkedIn, and the owned site. In legal and adjacent markets, the lists buyers already consult matter more than a generic guest post on a site nobody in the partnership has heard of. Mentions should point back to canonical pages so an assistant and a human land in the same place.

We refuse the shortcuts that keep resurfacing under new names: private blog networks, bought link packages, fake review volume, and “AI visibility” retainers that never change a page. Digital PR and Off-Site Visibility is the honest version of that work. Search and AI Visibility is where onsite and offsite sit together as one programme when the site can already be read.

Entity consistency sounds administrative. It is strategic. Models and humans both use repeated, stable descriptions to decide whether a firm is a real option or a vague category placeholder. If your LinkedIn says one thing, your directory listing another, and your practice page a third, you have taught the market three soft brands instead of one clear one.

How articles fit (and how they fail)

Service pages carry commercial intent. Answers carry citation shapes for questions buyers already ask. Insights argue a position with enough reasoning that a marketing lead can forward them to a partner. Confusing those jobs produces either a blog that never converts or a service page that reads like an essay.

Passage design is the editorial consequence of assistant retrieval. Put the answer in the first sentence of a section. Write sections that still make sense if lifted alone. Link deliberately into live services, stages, sectors, case studies and Pricing, not as a sprinkle of keywords. Refresh a living URL when facts move. Do not mint a new address for every rumour about an algorithm. Freshness on the page that already ranks beats a graveyard of thin posts.

A calendar of short articles will not repair a practice page that never says what happens next. We will say that in the first conversation if it is true for your site. The Visibility stage assumes the foundations underneath can hold the weight.

What we know about AI shaping SEO (and what we do not)

Enough is evidenced to act on. Assistants and generative answers favour clear, attributable, crawlable text. Content that only appears after client-side rendering is often missing from that layer. Measurement of what assistants fetch lives in server logs more than in the analytics suite. Zero-click behaviour already shrank what dashboards see. Pretending otherwise is how partnerships overfund the wrong channel.

What remains uncertain should stay labelled as uncertain. Exact weights inside closed models are not a public API. No single “AI SEO tool” reliably predicts citation. Brand can carry a weak page for a while, and nobody should sell certainty about how long. We optimise for human decision and machine quotation on the same sentences. We do not sell a secret prompt trick.

Practical consequence for the firm: stop waiting for a perfect map of the model. Fix the pages that carry demand, make the claims quotable, align the offsite descriptions, and measure with logs and enquiry quality beside sessions. That sequence is Foundations where the site cannot yet be read, then visibility, then growth spend.

Are AI-written articles a problem?

The problem is not that a model touched a draft. The problem is undifferentiated, unowned, inaccurate or unquotable copy shipped at volume.

AI drafting helps when it compresses research notes, inventories rival pages, or proposes a structure that a human then rewrites under a real brief. It fails when it invents case studies or figures, hedges until nothing is citeable, publishes ten near-duplicates that cannibalise one demand page, or puts regulated claims on the site without a human signer. Professional services buyers and assistants both punish empty fluency.

There is a second failure mode that looks respectable in a dashboard. A firm ships a high volume of “thought leadership” that never links to a live service, never states a refusal, and never answers the operational questions a buyer asks before instructing. Sessions rise. Enquiries do not. Assistants may ignore the posts entirely because nothing in them is specific enough to name. The marketing report calls the programme a success. The partnership feels the gap as “SEO does not work”.

What rules apply when a model is in the pipeline

When a model is in our pipeline, a short set of rules keeps the work defensible:

• A human owns the brief and the final.
• No invented proof, metrics or clients.
• Every insight has a named author who would defend it in a partner meeting.
• AI is a tool in the pipeline, not the byline. If partners will not put a name on the copy, the copy does not ship.

That discipline is slower than a content factory. It is also the only version that survives a compliance conversation and still gives an assistant something worth quoting. For tooling questions that sit beside the public site (what staff may paste into a public model, what evidence trail you need), the adjacent work is AI Governance and Assurance, not a blogging sprint.

What to take into the next partner meeting

If you remember nothing else from this piece, remember the three tests: findable, quotable, decidable. Rankings without the other two are an incomplete scoreboard.

Onsite, fix the commercial spine before you buy a new acronym. Put the deciding claims in the first HTML response. Treat technical hygiene as a floor. Rewrite demand pages before you scale paid media into them.

Offsite, make the firm’s name and language consistent, and make mentions point home. Refuse link schemes and empty AI visibility packs. Content should deepen a real buyer question or carry commercial intent. A blog calendar is not a strategy.

AI drafting is allowed as production help. It is not a licence to ship unowned volume. Named humans still own the brief, the facts and the byline.

If your firm’s rankings look healthy and the shortlist still leaves you out, start with whether the pages can be quoted and decided on. Start a conversation and we will say whether you are looking at a foundations problem, a visibility problem, or a conversion problem we should not take on.

Key takeaways

• Search widened into generative answers and assistants. It did not become a separate product you buy instead of SEO.
• The rewarded unit is increasingly a passage. Write sections that stand alone in ordinary HTML.
• Findable, quotable and decidable are three different tests. Rankings alone are not enough.
• Onsite spine and offsite corroboration still decide whether a professional services firm is a real option.
• AI drafting is a production tool. Undifferentiated volume without ownership is the ranking and reputation risk.
• Measure what assistants fetch as well as what humans click, then spend against pages that can convert.