Googlebot executes JavaScript. The crawlers that feed ChatGPT, Claude and Perplexity do not. Every technical SEO decision you make in 2026 sits downstream of that one asymmetry, and most scopes of work are still written as though there were a single crawler to satisfy.
This is not a small difference in degree. It is a difference in kind. A page that renders its main content client-side is a complete page to Google and a near-empty one to an AI crawler, which means the same URL can rank perfectly well and be structurally incapable of being quoted. Teams find this out late, usually after a quarter of content investment produces rankings and no citations.
So the scope question matters more than the vendor question. Below is what a technical engagement covers now, which findings rest on settled engineering and which rest on one two-year-old measurement, and where the money is actually worth spending.
What Do Technical SEO Services Actually Include In 2026?
Technical SEO services are an engineering engagement that makes a site retrievable, parseable and quotable by machines: crawl access, indexing, server-side rendering, status code and redirect hygiene, site architecture, internal linking, structured data, page performance and, since 2025, explicit access rules for AI crawlers. The deliverable is merged code, not a report.
Five workstreams make up a credible scope. They are listed here in the order a competent team sequences them, because each one makes the next measurable. Running structured data work before you have fixed rendering is how agencies produce a lot of activity and no movement.
| WORKSTREAM | WHAT IT FIXES | WHO OWNS THE FIX | MEASURABLE IN |
|---|---|---|---|
| Crawl access and budget | robots.txt rules, AI crawler directives, crawl traps, faceted URL explosion, server response under load | Agency specifies, platform or DevOps ships | Server logs, Crawl stats report |
| Rendering and content delivery | Main content present in server-delivered HTML rather than injected client-side | Front-end engineering, with the agency defining the acceptance test | View source, fetch without JS enabled |
| Status codes and URL stability | Redirect chains, soft 404s, stale URLs still being requested by AI crawlers months after a migration | Agency audits, platform ships | Server logs, Page indexing report |
| Architecture and internal linking | Orphan pages, depth, hub and spoke coverage, anchor consistency across templates | Shared: agency maps it, content and engineering implement | Crawl depth distribution, internal link counts |
| Structured data and machine readability | Schema that matches on-page facts, parseable tables, stable headings, no gated proof | Agency specifies, engineering ships | Rich Results Test, retrieval spot checks |
Notice what is missing from that table. There is no row for an AI visibility score, because no technical change has a defensible conversion rate into citations. We argued the general version of this when we called retrieval access a gate rather than a dial, and it applies here with force: fixing rendering does not buy you citations, it buys you eligibility for them.
Why Technical SEO Services Now Scope For Two Crawler Types
Two crawler populations now read your site with materially different capabilities, and a scope written for Googlebot alone will pass its own tests while leaving the AI surface broken. Googlebot renders, retries, and reports. The AI crawlers mostly do none of the three.
The measurement everyone cites for this is Vercel's crawler analysis with MERJ, published on December 17, 2024, built on traffic to nextjs.org and the Vercel network plus two job board sites as a consistency check. It is worth being precise about what it found, because the secondhand versions overstate it in one direction and underrate it in another.
| CRAWLER | EXECUTES JAVASCRIPT | 404 SHARE OF FETCHES | MONTHLY FETCHES OBSERVED | SCOPE IMPLICATION |
|---|---|---|---|---|
| Googlebot | Yes | 8.22% | 4.5 billion | Client-side content is readable, URL inventory is current |
| GPTBot and ChatGPT crawlers | No | 34.82% | 569 million | Needs server-rendered HTML and aggressive redirect hygiene |
| ClaudeBot | No | 34.16% | 370 million | Same, and it requested JavaScript files for 23.84 percent of fetches without running them |
| PerplexityBot | No | Not broken out | Not broken out | Server-rendered HTML only |
| Gemini | Yes | Not broken out | Not broken out | Rides Googlebot infrastructure, so Google-side fixes carry over |
| Applebot | Yes | Not broken out | Not broken out | Browser-based crawler, renders like a search engine |
The 404 column is the finding buyers underweight, and it is the cheapest thing on this page to act on. A third of what the ChatGPT and Claude crawlers fetched was a dead URL, four times Googlebot's rate. Those crawlers are working from stale inventories assembled at training and retrieval time, and they do not re-discover your sitemap on your schedule.
That turns two unglamorous jobs into high-yield work. Keep redirects resolving in one hop and keep them in place for years rather than months, because the requesting population has a much longer memory of your old URLs than Google does. We went through the mechanics of this in detail when we looked at why AI crawlers cannot see client-rendered pages.
“Googlebot tells you when it cannot read your page. The AI crawlers fail silently, and the first signal you get is a competitor being quoted instead of you.”
Be fair about the age of this evidence, because it cuts both ways. The Vercel figures are from December 2024, which in this market is a long time, and no comparable large-scale re-test has been published since. Treat the direction as reliable and the exact percentages as dated. Then do the thing almost nobody does and verify it on your own property: request your top twenty pages without executing JavaScript and read what comes back.
Which Findings Are Settled Engineering And Which Are Contested?
A technical finding is settled when a named, reproducible mechanism connects the fix to the outcome, and contested when the connection rests on correlation across datasets that disagree with each other. Both belong in a scope. They do not belong at the same confidence, and they should not be priced the same way.
The contested half is contested for a reason you can check yourself. Published estimates of how many AI citations also rank in Google's top 10 run from 12 percent to 84 percent across five credible research teams, a spread we worked through when writing about what SEO audit services can honestly measure. Profound's February 2026 research put the share of AI citations explained by standard SEO ranking factors at 4 to 7 percent.
Confidence a buyer should assign to each technical workstream, scored on whether a named reproducible mechanism links the fix to the result. Illustrative scoring, based on the evidence quality discussed above
Read the bottom two bars as a buying rule. Structured data is genuinely useful for machine readability and genuinely oversold as a citation mechanism. A composite AI visibility score, computed by a vendor from a method it will not publish, is not a measurement you can act on or audit, and a proposal that leads with one is telling you how it intends to report success.
The top three bars are where a fixed-scope engagement earns its fee. They are diagnosable in days, the fixes are unambiguous, and the result is verifiable by someone other than the agency that did the work. That verifiability is the whole point, and it is the standard we hold our own SEO and GEO audits to before anything gets prioritised.
How Much Of This Should Your Own Engineers Keep?
Most of the implementation should stay in house, and the agency should be buying you specification, prioritisation and acceptance tests rather than commits. The exception is diagnosis, which is where outside pattern recognition is worth real money because your team sees one site and a good technical team sees fifty.
The split above also tells you what a bad proposal looks like. If the bulk of the hours sit in implementation your team could do, and the diagnosis is a tool export with the agency logo on the cover, you are paying consultancy rates for throughput. The reverse shape, heavy on diagnosis and specification with your engineers shipping, is where the economics work.
How To Buy Technical SEO Services Without Buying A Crawl Report
Buying well comes down to five questions asked before signature, each of which a weak vendor answers with a tool name and a strong one answers with a method. Ask them in this order, because the first answer tells you how seriously to take the rest.
One caution about sequencing, because it is the most common expensive mistake. Teams arriving from a content programme want the structured data and the entity work first, since that is the part that sounds like it produces citations. If rendering and status codes are broken underneath, that work lands on pages the AI crawlers cannot read, and you will conclude that structured data does not work when what actually happened is that nothing you shipped was ever retrieved. We built a dependency framework for this sequencing for exactly that reason.
It is also worth saying what has not changed, because the 2026 framing can make teams abandon things that still pay. Google is still the larger source of qualified traffic for almost every B2B site we work on, the November 2026 core update is rolling out right now, and Search Console still reports impressions without clicks on its AI surfaces. Classical technical SEO did not stop mattering when a second crawler population arrived. It got a second audience with stricter requirements.
Common Questions About Technical SEO Services
The questions below are the ones that come up in every scoping call, answered short enough to forward to an engineering lead who has no interest in search terminology.
See where you are cited today
A free snapshot audit of your rankings and AI citations before we ever talk.
Josh leads work at the intersection of SEO and generative engines at Something Inc., helping B2B brands get ranked and cited across every major AI engine.