Technical SEOSeptember 9, 202612 min readBy the Kiwi senior team

Technical SEO is the platform AI search runs on — and most of the AI-SEO budget is unverifiable

69% of teams have allocated budget to AI-search optimization, but 80% report 'no idea' or under 5% of revenue. Meanwhile the Core Web Vitals that actually gate visibility stay broken. The checked 2026 data on what's load-bearing and what's a retainer without a denominator.

In this guide

The budget is allocated, the visibility is tracked, and the revenue is unknown
Crawlability is the real gate — and it is technical, not tactical
The three metrics that still gate the experience
What the 2026 field data actually shows
INP is the metric nobody is fixing — and it is the rising one
The honest counter-evidence: the revenue story is real but narrow
The playbook that is provable end to end
Questions
Sources

The budget is allocated, the visibility is tracked, and the revenue is unknown

The SEOFOMO State of AI Search Optimization report for 2026 is the most useful survey this year because it measures the gap between spending and proof. Sixty-nine percent of respondents have allocated budget to AI search, 75% have a dedicated strategy, and 92% track visibility. But only 51% see a positive impact, and 80% report either no idea of the revenue effect or a share under 5% of tracked revenue. The industry has moved into AI search ahead of its own measurement.

Read those two columns together: almost everyone is spending and watching, but only half reports an effect and four in five cannot tie the spend to revenue. That is the honest starting point, and it explains the rest of what follows. The load-bearing work — the technical foundation that determines whether an AI engine can crawl, parse, and cite you at all — is measurable and real. The 'AI SEO' layer sold on top of it is, in most cases, a retainer without a denominator.

69% have allocated AI-search budget; 75% have a strategy; 92% track visibility (SEOFOMO 2026).
Only 51% see a positive impact; 80% report 'no idea' or under 5% of revenue.
The market is spending ahead of its measurement.

Crawlability is the real gate — and it is technical, not tactical

An AI engine can only cite what it can crawl, parse, and index. The technical foundation is the thin end of the funnel for every other layer, and it is where the evidence is clearest. The 2026 AI Visibility pricing stacks sold by agencies are explicit about this on their own websites: the 'Layer 1 technical foundation' — robots.txt directives, XML sitemaps, IndexNow setup, schema deployment — is priced at $2,000 to $2,500 one-time or $750 a month ongoing, before any content, entity, or 'LLM seeding' work starts at $2,500 to $8,000 a month on top of it.

The point is not that agencies are dishonest. It is that the same stack that positions 'LLM seeding' and 'prompt auditing' at the top advertises its own bottom layer as plain technical SEO. Ninety-two percent of teams already track visibility, which means the winning move is not another monitoring dashboard — it is making your pages actually resolve for the crawlers that feed every engine. No amount of citation outreach compensates for a site the crawler cannot reach.

AI SEO 'Layer 1' is robots.txt, sitemaps, IndexNow, schema — plain technical SEO, priced at $750/mo.
Content, entity, and LLM-seeding layers stack on top at $2,500–$8,000/mo.
Crawlability is the gate: an engine cites what it can parse, not what the retainer claims.

The three metrics that still gate the experience

Core Web Vitals are the one part of technical SEO with public, fixed thresholds and a confirmed ranking role. LCP measures loading and must be 2.5 seconds or less; INP measures responsiveness and must be 200 milliseconds or less; CLS measures visual stability and must be 0.1 or less. All three are assessed at the 75th percentile of real visits, on mobile and desktop separately, and a page passes only when all three are good simultaneously.

They have been a page-experience ranking input since the 2021 update — a tiebreaker and threshold rather than a primary factor, which is why a slower page with stronger content can still rank. The practical effect is competitive and indirect: a similar competitor that passes gains an edge, and poor scores raise bounce, which hurts rankings through a second-order effect. Treat them as the floor the AI-era visibility stack stands on, not as the ceiling.

LCP ≤ 2.5s, INP ≤ 200ms, CLS ≤ 0.1 — assessed at p75, mobile and desktop separately.
A confirmed ranking input since 2021, but a tiebreaker/threshold, not a primary factor.
Clearing the floor is necessary; it is not the differentiator the layers above claim.

What the 2026 field data actually shows

The gap between vendor benchmarks and field reality is the recurring theme of 2026 performance data. Shopify's own theme table shows median pass rates in the 80–97% range for LCP, INP, and CLS — a signal that everything is fine. A real multi-page lab sample of 64 live Shopify stores flips the story: mobile Lighthouse Performance averages cluster near 51 out of 100, median mobile LCP is 9.8 seconds, and exactly zero of the 64 store averages cleared the 2.5-second LCP threshold. The famous DTC names tell the same tale — Fashion Nova, SKIMS, Brooklinen, Kylie Cosmetics all show mobile LCP averages between 8.6 and 31.1 seconds in lab testing.

Field data (Chrome UX Report) is kinder but still sobering. In a September 2026 CrUX sample at origin level, Allbirds passed all three Core Web Vitals, but the rest of the cohort — ColourPop, Kylie Cosmetics, Netflix Shop, Steve Madden, Glossier, Fashion Nova, SKIMS, Brooklinen — sat in 'needs improvement' or 'poor,' almost always on INP and CLS rather than LCP. The pattern across the whole dataset is consistent: CLS is usually fine, LCP is the structural problem on heavy hero images, and INP is the rising problem as sites bolt on more third-party scripts and chat widgets. Layout is already handled; loading and responsiveness are not.

Shopify theme table: median 80–97% pass. Real 64-store sample: median mobile LCP 9.8s, 0/64 clear 2.5s.
DTC names: Fashion Nova 8.6s, SKIMS 9.1s, Brooklinen 31.1s, Kylie 26.8s mobile LCP (lab).
CLS is the bright spot; LCP breaks on hero images, INP breaks on third-party script bloat.

INP is the metric nobody is fixing — and it is the rising one

Interaction to Next Paint replaced First Input Delay in 2024, and in 2026 it is the hardest Core Web Vital to pass on Shopify specifically and on JavaScript-heavy sites generally. Median mobile INP in the 64-store sample is 300ms — over the 200ms good threshold — and it is almost always an app-and-theme JavaScript problem rather than a hosting problem. Long tasks on the main thread delay the next paint after every tap, form submission, and menu open.

The fix is invisible but mechanical: audit and remove unused apps and embeds, split long tasks with scheduler.yield or setTimeout, move heavy work to a Web Worker, defer third-party scripts, and debounce expensive interactions. Because Google evaluates field data over a rolling 28-day window, the metric lags the fix by weeks — which is why teams that declare failure after a weekend of work misread a slow measurement loop as a failed intervention. Hold the deploy, then hold the nerve for the field window.

INP replaced FID in 2024; median mobile INP 300ms in the 64-store sample (200ms is the good line).
Almost always app/theme JavaScript — long main-thread tasks — not hosting.
Field data is a 28-day rolling window: the metric lags the fix. Verify in CrUX, not the lab.

The honest counter-evidence: the revenue story is real but narrow

The measured commercial case for speed is real and worth stating without inflation. A Vodafone A/B test on two functionally identical pages found that a 31% LCP improvement drove an 8% sales lift — a direct revenue variable in a controlled comparison, not a correlation. Google's research consistently associates each additional second of load time with a 7–11% drop in conversions, and a 0.1-second mobile improvement with improved retail outcomes. These are directional, context-dependent numbers, and the source itself warns the effect depends on theme, scripts, product type, and traffic mix.

But that is the honest limit of the commercial story, and it is exactly why the AI-SEO layer above technical SEO is in the position the SEOFOMO survey describes: 51% positive impact, 80% unable to measure revenue. Technical SEO has a known denominator — a page either resolves for a crawler or it does not, a metric either clears a threshold or it does not, and a controlled test either lifts sales or it does not. The layers priced above it trade in visibility scores without an auditable revenue link. The teams coming out of 2026 ahead are the ones spending the marginal dollar on the foundation they can prove, not the retainer they cannot.

Vodafone A/B: 31% LCP improvement drove 8% sales lift — a controlled, causal result.
Google: ~7–11% conversion drop per additional second of load time — directional, context-dependent.
Technical SEO has a known denominator; the layers above it mostly do not.

The playbook that is provable end to end

Instrument before you optimize. Record the mobile and desktop 75th percentile for LCP, INP, and CLS per page template from Search Console, PageSpeed Insights field data, and your own RUM, and decide in advance whether you report origin-level, URL-level, or lab numbers — never mix the three in one column. Fix in the order the data points: LCP first (serve the hero as WebP or AVIF, preload with a high fetchpriority, never lazy-load it, strip render-blocking CSS and JavaScript), then INP (the app-and-theme JavaScript audit), then CLS only if it is actually broken, which it usually is not.

Then hold the line on the measurement. Set a regression alert on the Search Console Core Web Vitals report, re-test after every app install, theme update, and campaign, and treat performance as a governed property rather than a one-off project — scores drift back the moment you stop watching. On the AI-search side, keep the foundation first: confirm every AI-relevant crawler can reach and parse the pages that matter, and spend the visibility budget only where it carries a denominator you can audit against a holdout. Technical SEO earns its place because you can prove it. That is the entire reason it is the platform and the rest is a layer on top.

Baseline p75 LCP/INP/CLS per template from field data; never mix origin, URL, and lab numbers.
Fix LCP first, INP second, CLS only if broken.
Set CWV regression alerts; re-test after every app/theme/script change.
Keep the crawler-access foundation first; spend visibility budget only against an auditable denominator.

Frequently asked questions

Is AI-search optimization actually worth spending on in 2026?

The evidence says the foundation is worth it and the layer above is unproven. SEOFOMO's 2026 survey found 69% have allocated budget but 80% report no idea of the effect or under 5% of revenue. The technical foundation — crawlability, parseability, and Core Web Vitals — has a known, provable denominator. The 'LLM seeding' and 'prompt auditing' layers above it mostly trade in visibility scores with no auditable revenue link. Spend the marginal dollar on the foundation you can prove.

Are Core Web Vitals still a Google ranking factor in 2026?

Yes, with an important qualification. They have been a page-experience ranking input since the 2021 update, but a tiebreaker and threshold rather than a primary factor — a slower page with stronger content can still rank. The real effect is competitive and indirect: a passing competitor gains an edge, and poor scores raise bounce, which hurts rankings second-order. Treat them as the floor the visibility stack stands on.

Which Core Web Vital is hardest to pass in 2026?

INP, the responsiveness metric that replaced First Input Delay. It is almost always an app-and-theme JavaScript problem — long main-thread tasks — rather than hosting. The 64-store Shopify sample shows median mobile INP at 300ms against a 200ms good threshold. Fix it by removing unused apps, splitting long tasks, moving heavy work to a Web Worker, and deferring third-party scripts.

Why do lab scores and field scores disagree so much?

They measure different things at different stages. Lab data (Lighthouse) runs on throttled, synthetic conditions and catches LCP candidates, which is why the 64-store sample shows 8–31 second mobile LCP. Field data (Chrome UX Report) reflects real users at the 75th percentile over a 28-day rolling window, and is what Google actually ranks on. The same site can show a 34 lab Performance score while passing all three Core Web Vitals in field data, as Allbirds did.

How do I make technical SEO provable rather than a retainer?

Instrument first: baseline p75 LCP, INP, and CLS per template from field data, and fix in data order — LCP, then INP, then CLS only if broken. Confirm every AI-relevant crawler can reach and parse the pages that matter. Set regression alerts and re-test after every app, theme, and script change. On the AI-search side, require any tool or agency to name a denominator you can audit against a holdout — a page either resolves or it does not, and a claim either survives a control or it does not.

Sources

Figures cited above are drawn from the linked publications and are the responsibility of their sources; we date and scope them rather than presenting them as universal guarantees.