AI Deployment
Feature 32  ·  Synthesis  ·  Edition Q1 2026

The index was
always the asset.

Five features on this desk describe the same object from different sides. A court ordered it shared. Two-fifths of top sites have withdrawn from it. One site we audited was never in it. What SEO optimised for twenty years was position within an index; what now decides outcomes is presence in one - and those turn out to be very different problems.

Search marketing spent two decades on a question of order. Given that everyone is in the index, who appears first.

Every technique in the discipline assumes that premise. Keyword targeting, internal linking, backlink acquisition, page speed - all of it operates on relative position among documents that are already present and already readable.

This desk documents the premise coming apart from four directions at once.

Figure 01
Four ways presence became the question
The desk, reduced to one variable.
27
A court ordered the index shared. The remedy that matters is not the ban on default contracts - it is index access and syndication. The asset that made search a single-destination market is the corpus, and a court has said others may have it. In force since 3 March 2026, with the terms still unwritten.
28
Two-fifths of the top thousand withdrew. 40.9% unreadable to GPTBot; 82.4% of news sites restricting at least one crawler. Not a ranking decision. A presence decision, taken deliberately, in a file most organisations have never read.
30
One site was never present at all. Twenty-eight DOM nodes where three thousand words should have been. Correct metadata describing a document with no content. Not ranked badly - not a candidate.
29
There is now more than one index worth being in. Google at 89.87% of conventional search, ChatGPT at roughly two-thirds of chatbots, Perplexity growing at 370%. Presence is no longer a single binary but a set of them.
This reduction is ours. Each underlying figure carries its original grading from the feature it comes from. The argument that these are one problem rather than four is an editorial claim, not a finding any source makes.

Rank asks where you sit among documents that are present. Every finding on this desk is about whether you are present at all.

Figure 01
The state of the index
Three published measures of how much of the web is now withheld from, or bypassed by, the crawling layer.
Bar chart: 40.9% of top sites blocking GPTBot, 82.4% of news sites restricting at least one AI crawler, 38% of AI Overview citations from the top ten results.
Source: figures as published in Features 22, 28 and 32 of this edition. Nothing here is newly calculated.

Why the distinction was invisible before

Presence used to be free and automatic. You published a page, a crawler found it, and it entered the index. The only interesting question was position, so the discipline built itself entirely around position - and the vocabulary followed. Rankings. Positions. Competitors. Share of voice. Every term describes relative order.

Three things made presence a decision rather than a default.

Crawling became contested. Feature 28's split between training and search crawlers means access is no longer one permission but several, each with different returns and each grantable separately.

Rendering became conditional. A document served as a shell is present as a URL and absent as content. Feature 30's case is extreme; the class is ordinary.

Retrieval replaced ranking for a growing share of answers. The Content desk's Feature 22 found four in five language-model citations coming from pages outside the top hundred. A retrieval system does not need you to rank. It needs you to exist, in readable form, containing an extractable claim.

What this implies for how the work is ordered

If presence is a precondition and position is a refinement, the sequence most organisations run is inverted.

The conventional order
Position first
1. Keyword and topic researchWhat to compete for.
2. Content productionPages against those topics.
3. Link acquisitionImproving relative position.
4. Technical auditUsually annual, often after a problem.
What the evidence supports
Presence first
1. Served-document checkDoes a crawler receive your content - Feature 30. Seconds to run, catastrophic when wrong.
2. Crawler policy decisionWhich systems may read you, deliberately chosen - Feature 28.
3. Citation baselineWhere you are named now, across the destinations that matter - Feature 29.
4. Then position workWhich remains necessary and has become insufficient.

The right-hand column is cheaper than the left. That is the part worth noticing. Steps one and two cost hours and require no budget approval; step three costs a baseline measurement. The expensive work is unchanged - it simply now sits behind three cheap checks that determine whether any of it can work.

The honest state of the discipline

Four things this desk could not establish, and they matter more than most of what it could.

Figure 02
What nobody has published
The gaps that would change the advice.
How common rendering failure actually is
Feature 30 documents one case in production on a well-built modern site. No study measures incidence across the web.Would determine whether this is a footnote or an epidemic
Whether crawler blocking helps or harms the blocker
40.9% of top sites have made a decision with no published evidence on either side of it.Would settle the largest open question in Feature 28
Whether the antitrust remedies will function
In force since March, terms unwritten, under appeal with no briefing schedule.Determines whether search is one destination or several
What any of this does to revenue
Citation, presence and eligibility are all measured in their own units. The Content desk found no research linking citation to commercial outcome, and this desk found none either.The question every reader actually has
These are absences of evidence, not evidence of absence. Each is a study somebody could run. We name them because a desk that reports only what is known overstates how much is known.

Whose interest this serves

This desk concludes that technical presence work - audits, crawler policy, rendering checks - has become the precondition for everything else. Marketing Legendary sells exactly that work. It is the most commercially convenient conclusion available from these findings. Two things offered against it: the three checks recommended above require no vendor and take hours, and Feature 31 proposes holding search functions to standards that are harder to satisfy than the traffic targets they currently miss. Neither makes the conclusion disinterested. Both make it checkable.

What survives every scenario

The uncomfortable feature of this desk is how much of it is unresolved. The remedies may be stayed. The blocking wave may reverse. The growth rates may decay. Any confident plan built on a single outcome here is a bet.

So it is worth asking which actions are correct across all of the plausible outcomes, because those are the ones to take now.

Correct in every scenario
No bet required
Serving your content in the documentNecessary for ranking, for retrieval, for reader modes, for every destination that exists or arrives.
A deliberate crawler policyWhatever you decide, deciding it beats inheriting a template.
Knowing where you are cited todayA baseline is worth more the earlier it is taken and costs the same whenever it happens.
Separating the two markets in reportingUseful whether one wins or both persist.
Requires a bet
Defer or size deliberately
Reallocating budget away from Google89.87% share says no in 2026. Growth rates say ask again.
Blocking training crawlersNo published evidence on either side. A reversible decision, so treat it as one.
Building for a multi-destination marketDepends entirely on remedies that are in force and unusable.
Optimising for citation shareThe Content desk found no link to revenue. Measure it; do not target it yet.

The left column is unglamorous, cheap and entirely within an organisation's control. The right column is where the strategic conversations happen and where almost all the attention goes.

That inversion is the desk's practical finding. The actions that are unambiguously correct are the ones nobody is discussing, and the ones being discussed all depend on outcomes nobody knows.

What to do about it

Run the presence checks before the next content brief. Served-document inspection, crawler policy review, citation baseline. All three this month, none requiring approval.

Stop reporting a single organic number. Conventional search and AI answering have different leaders, different trajectories and different mechanics. One line collapses them and hides which is moving.

Name the owner of every input in Feature 31's Figure 01. The exercise produces the escalation argument by itself.

Watch two dates on the antitrust case. Whether a stay is granted, and when the Technical Committee publishes licence terms. Neither will be reported by search marketing publications and both change the market's shape.

Assume presence is now a maintained property. It used to be a one-time achievement. A deployment can remove you from consideration without any visible symptom, which means the check belongs in the release process rather than in an annual audit.

Figure 03
The SEO desk, Q1 2026
Every headline figure from the six features.
FindingValueFeature
Market structure
Remedies decision2 Sept 202527
Chrome divestitureRejected27
Data-sharing regime triggered3 Mar 202627
Licence terms finalisedNo27
Google appeal filed20 Jan 202627
Presence
Top 1,000 sites unreadable to GPTBot40.9%28
News sites restricting an AI crawler82.4%28
DOM nodes served, audited site2830
Where to be present
Google conventional search share89.87%29
Google share of referrals87.6%29
ChatGPT chatbot share64.5–68%29
Perplexity user growth+370%29
Perplexity referral share vs Gemini7.67% vs 7.03%29
Not established anywhere on this desk
Incidence of rendering failure-No study
Whether blocking helps the blocker-No study
Whether the remedies will function-Unresolved
Revenue effect of any of it-No research
Assembled from separate sources with different samples, periods and methods. The final group is the honest state of the evidence.

How we did this

Where this comes from
Synthesis. Every figure is carried from Features 27–31 of this desk and Features 21–23 of the Content desk, with original grading intact. No new external source is introduced.
What's ours, not the source's
The position-versus-presence reduction, the inverted sequence and the four named gaps are entirely ours.
Interest
Disclosed prominently. This is the most commercially convenient conclusion the desk could reach and we say so in the text.

What this doesn't prove

  • That presence work improves outcomes. It is a necessary condition, which is a weaker claim than sufficiency and the only one the evidence supports.
  • That the four findings are one problem. The reduction is an editorial argument. They could equally be four unrelated changes arriving together.
  • That position work has stopped mattering. Google still holds 89.87% of conventional search and ranking still supplies a fifth of LLM citations.
  • Any revenue effect. Named as the fourth gap. Nothing on this desk or the Content desk connects any of these measures to a commercial outcome.
  • That the recommended sequence outperforms the conventional one. Untested. It is ordered by cost and by necessity, not by demonstrated return.
  • Anything about markets outside those measured. Share and blocking data are heavily US and English-language weighted.

Sources for this feature

  1. Features 27, 28, 29, 30 and 31 of this desk. Another feature in this edition - all external sourcing and grading carried from those features
  2. Features 21, 22 and 23 of the Content desk. Another feature in this edition
End of the SEO desk
Q1 2026 · Six Features
Marketing Legendary publishes quarterly. The next edition follows.
SE
The practice behind this desk

SEO Legendary

We check what a crawler actually receives before we discuss rankings, because the two are not the same document.