The Technical Signals AI Search Uses That Most SEOs Still Aren’t Optimizing

Sep 02, 2026 02:00 AM - 1 hour ago 3

For the past mates of years, the taxable of AI visibility has dominated SEO discussions. SEO teams person cleaned codification and chunked content, each the easier for AI bots to entree and ingest. If your marque has started earning regular citations successful AI-generated answers aliases showing up successful AI Overviews, you mightiness beryllium forgiven for reasoning the conflict for AI visibility is astir won.

None of that is wrong, and nary of it is wasted effort. It’s just not the full story.

However, our audit of 50 awesome websites recovered that, while astir person made it easier for AI to find them, almost nary person made it imaginable for AI to truly understand them. And nearly two-thirds leave the mobility of which AI bots tin entree which contented wholly to luck.

3 Layers Of AI Visibility

Ask a trading squad to specify “AI visibility,” and the reply will apt attraction connected mentions and citations. Should personification inquire immoderate of the starring AI platforms a mobility applicable to your category, you want your marque and/or merchandise to person a amended than mean chance of appearing successful the response.

Defining AI visibility purely successful position of citations and mentions is aged SEO reasoning – get ranked, get found, get clicked – carried complete to a caller and very different shape of search. But visibility successful AI isn’t simply astir getting your brand, merchandise and applicable links successful beforehand of the correct eyeballs. It’s besides astir really AI understands your accusation and interacts pinch your website.

If you’ve parented a kid done superior school, you already cognize that school them to publication isn’t the aforesaid arsenic school them to understand. A six-year-old learning phonics mightiness publication the words successful their early reference book reasonably fluently. Tick, get yourself a cooky from the jar. But inquire them astir what they’ve conscionable read, and you whitethorn find they absorbed aliases understood little than you thought.

Once a kid has mastered some reference and comprehension, they tin put what they study into action – travel a look to navigator a meal, constitute their ain stories aliases essays, transportation retired further investigation to find answers to questions arising from the first text, and truthful on.

Reading is only 1 measurement towards comprehension. And comprehension is only 1 measurement towards agency; what you determine to do pinch that accusation erstwhile you person it.

This is why we developed an audit model for AI readiness pinch 3 chopped layers:

1. Retrievability

Can AI fetch and parse your contented without stumbling? This is the reference bit, the foundational furniture astir SEO teams already optimize for, straight related to AI visibility successful the consciousness of citations and marque mentions.

But unless you besides optimize for the 2nd layer, you’re fundamentally crossing your fingers and hoping AI interprets your contented correctly.

2. Attribution And Meaning

Can AI find what your pages are astir and who owns them? This is really AI knows, without simply guessing, which number connected the page is the merchandise price, and which is the discount value disposable exclusively to loyalty paper holders. It’s really AI determines who the marque and/or writer really is, and whether to see them a trusted authority connected the subject. It’s really AI tells whether an article is astir Jaguar the car brand, aliases 1 of the galore Jaguar shot clubs.

Giving AI greater assurance that it has correctly interpreted your contented does 2 things: It increases the likelihood of your contented being cited successful applicable AI responses, while besides reducing the consequence of those responses misrepresenting your brand, your products aliases your claims.

3. Agent Transaction And Discovery

Can AI agents entree your website’s capabilities to transportation retired tasks, specified arsenic completing transactions? This last furniture is astir agency, and it’s the quality betwixt AI simply parroting accusation backmost to the personification and being capable to interact pinch your business connected their behalf.

Under each of these layers, we grouped the apt protocols SEO teams mightiness instrumentality to achieve those outcomes. For example, implementing ARIA labeling aliases including AI user-agent directives successful the robots.txt would mostly subordinate to Retrievability, while the beingness of an entity map aliases JSON-LD schema would subordinate to Attribution and Meaning.

Altogether, our model includes 27 audit elements – 11 successful Layer One, 3 successful Layer Two, and 13 successful Layer Three – rated and scored according to their existent maturity arsenic manufacture standards:

Established: Production-ready standards successful progressive usage today.
Emerging: Real protocols gaining traction pinch early movers, and increasing fast.
Frontier: Standards still being debated pinch nary settled implementation truthful far.

Armed pinch this framework, we audited 50 awesome websites crossed retail, SaaS, travel, publishing and finance to spot wherever the astir communal gaps successful AI readiness mightiness be.

I’ll admit it’s a tad unfair of maine to use the aforesaid model and scoring strategy crossed the full cohort, sloppy of manufacture and business model. When utilizing our AI readiness model pinch clients, we weight the scoring according to their industry. But utilizing the model arsenic a benchmarking instrumentality crossed a varied cohort wouldn’t activity if we measured different things aliases applied different scoring to each website.

The intent was to item wherever the biggest gaps are overall, alternatively than guessing astatine the nuances of each individual business lawsuit aliases commercialized decision. A debased people doesn’t needfully mean a tract is underprepared for AI if different clues propose they’ve deliberately adopted this approach.

See also: Machine-First Architecture: How To Build Websites Machines Can Identify, Read, Cite & Use

What The Data Shows

Using an instrumented browser, we were capable to seizure unrecorded HTTP responses, the rendered DOM, earthy server HTML, and machine-discovery endpoints. All information was captured connected the aforesaid day: June 12, 2026.

We past scored 12 Established signals against our agentic readiness model (0/1/2), pinch the last people rendered arsenic a percent of the imaginable maximum. While we besides tracked Emerging and Frontier protocols, we excluded these from the scoring.

The highest people went to Airbnb.com pinch 79.2%.
On average, sites person implemented conscionable complete half of the established protocols (mean 56.6%, median 58.3%).
Only 10 sites scored beneath 50%.

While that mightiness sound for illustration a walk mark, it isn’t. When you propulsion the 3 layers apart, it becomes clear wherever there’s still activity to beryllium done.

1. Retrievability (Average 74.4%)

Established Audit Elements (8/11)

txt & AI User-Agent Directives.
Accessibility Tree Integrity.
ARIA Labeling & Descriptive Names.
Semantic HTML & Document Hierarchy.
Token-Efficient DOM Density.
Server-Rendered / Clean HTML Delivery.
Form & Input Machine Usability.
Sitemap Declaration.

This is the furniture that overlaps astir heavy pinch accepted method SEO, including elements that whitethorn person already been successful place, making immoderate caller tweaks to optimize for AI retrievability easier to implement. Unsurprisingly, astir sites execute reasonably well, pinch only 3 scoring beneath 50%. And location could beryllium morganatic reasons for those debased scores, arsenic you’ll see.

2. Attribution And Meaning (Average 38.5%)

Established Audit Elements (2/3)

JSON-LD Schema & Semantic Richness.
Content Signals Policy.

Scores driblet distant sharply betwixt the first and 2nd layers.

The bully news is we detected JSON-LD system information connected the homepage of 35 retired of 50 websites (70%), pinch each but 3 scoring the 2-point maximum. The bad news is that still leaves astir a 3rd without immoderate astatine all.

Schema is simply a captious portion of really AI understands what your contented really means: This is simply a product, this is the price, and this is the marque that sells it. Poor aliases non-existent schema astir apt won’t undermine your website’s AI retrievability, and the LLM will still make its champion conjecture astatine what everything means. But that champion conjecture tin besides beryllium wrong, starring to inaccurate AI responses aliases moreover outright hallucinations.

Perhaps much telling is that only 5 of the 50 audited sites person implemented Cloudflare’s Content Signals Policy. This is simply a group of directives wrong the robots.txt record that spell retired precisely what crawlers are permitted to do pinch your contented successful narration to hunt indexing, unrecorded AI query responses, and exemplary training. A Content Signals Policy intends your AI strategy isn’t constricted to a simplistic binary prime – “block each AI” aliases “allow each AI” – allowing you to return a much strategical approach.

3. Agent Transaction And Discovery (Average 2.1%)

Established Audit Elements (2/13)

OAuth Discovery (Authorization Server).
OAuth Protected Resource Metadata.

Agentic browsers and agentic commerce person only been astir since precocious 2025. As a result, of the 13 protocols we identified arsenic straight related to Layer 3, we could only people 2 arsenic Established.

Both ebay.com and wikipedia.org artifact same-origin fetch via CSP. But of the 48 sites wherever endpoint testing was possible, 46 scored zero. And while airbnb.com and vercel.com person some implemented OAuth authorization server metadata, neither has implemented OAuth protected assets metadata, meaning they only scored 50% each for Layer 3.

OAuth find and OAuth protected metadata make it imaginable for AI to interact pinch your site. For example, the OAuth find metadata tells a customer exertion what it’s allowed to access, really to place itself, and wherever to get the accusation it needs to securely entree an OAuth authorization server aliases protected API.

However, our squad is intimately monitoring 9 different Emerging elements successful Layer 3 that could spot the abstraction germinate quickly.

Some subordinate to the Model Context Protocol (MCP), which allows AI to query your servers directly. For example, alternatively of piecing together accusation from your website merchandise pages, AI gets what it needs consecutive from your real-time merchandise database, drastically reducing processing clip and the consequence of AI hallucination.

And past location are the comparatively caller ecommerce protocols that let AI to complete transactions wrong AI conversations, for illustration Google’s Universal Commerce Protocol (UCP) and OpenAI’s Agentic Commerce Protocol (ACP), which are apt to go established successful the adjacent future.

Layer 3 is very overmuch the area to watch, pinch the imaginable to confer first-mover advantage connected immoderate brands consenting to experiment.

A Low Score Isn’t Always A Bad Score

I’m not suggesting each business should clasp AI successful the aforesaid way, pursuing our model of established protocols for illustration a checklist to beryllium completed. While it’s technically imaginable for a website to people 100%, that doesn’t mean it should.

Some publishers wrong the cohort – including the BBC, CNN and The Guardian – specifically artifact most, if not all, AI bots. This is understandable. Their business models dangle connected group visiting their sites to get the latest news and publication their peculiar marque of reporting.

Meanwhile, industries specified arsenic retail, SaaS, recreation and financial services are progressively limited connected reaching customers and influencing decisions successful these AI spaces. And arsenic much customers commencement interacting pinch these businesses via AI agents alternatively of done their websites, each 3 layers will go progressively important.

Then again, 1 of the lowest wide scores (29.2%) belongs to amazon.com, which was besides 1 of the 3 sites to people beneath 50% successful Layer 1. Now, I don’t deliberation for 1 infinitesimal that Amazon has simply overlooked AI search. Like the BBC, Amazon has circumstantial AI crawler directives successful spot to artifact almost each AI bots from accessing the site, truthful this is intelligibly by design. And depending connected their reasons for blocking the bots, location mightiness not beryllium overmuch constituent successful implementing immoderate of the different measures.

Conversely, immoderate sites, for illustration airbnb.com and cloudflare.com, person nary general blocks successful spot but person implemented circumstantial entree rules for astir of the awesome AI bots. Others, including eBay.com and tripadvisor.com, person been moreover much selective, implementing rules for the AI bots they do want and blocking those they don’t.

Each of the supra sites has made deliberate, bot-specific decisions and updated their robots.txt pinch the due AI directives. They’ve begun to instrumentality a scheme for AI moreover if, successful immoderate cases, that scheme is to artifact it wholly – successful which lawsuit an AI readiness people is benignant of moot.

But astir two-thirds of the audited sites (29 of 50) look to person made nary deliberate determination astir AI supplier entree 1 measurement aliases the other. Nothing is blocked, and thing is explicitly allowed either, pinch nary rules for crawlers to follow.

What AI bots tin access, really they construe what they find, and really they usage that accusation successful AI responses is mostly near to chance.

Signals Are Not Guarantees

At this point, we should besides mention llms.txt, which goes beyond the AI directives and contented signals wrong robots.txt to supply AI systems pinch a curated, human-readable guideline to a site’s contented and structure.

Our model classifies llms.txt arsenic a Frontier element, excluded from scoring, because it is presently an unratified modular pinch nary agreed specification assemblage down it. However, we recovered 11 of the 50 sites person published one, suggesting there’s immoderate enthusiasm for the thought of providing AI pinch a useful “CliffsNotes” guideline to a marque and its content.

But earlier you instruct your squad to update your site’s robots.txt aliases adhd an llms.txt, it’s worthy clarifying what these files tin – and can’t – do.

Any AI directives documented successful your robots.txt are penchant statements, alternatively than hard-coded rules. Compliance is voluntary. The awesome AI crawlers – GPTBot, ClaudeBot, Google-Extended and Applebot-Extended – broadly respect them, but beyond that group, things get patchier. Some bots whitethorn crawl your contented regardless.

Content signals transportation moreover little enforcement weight. Well-behaved bots tin take to travel them, but there’s nary method system to unit compliance. And, for now astatine least, llms.txt is still small much than a awesome of intent.

For example, Expedia has published an llms.txt record that sounds for illustration a emotion missive to AI, mounting retired the brand’s canonical personality and explaining successful neatly chunked transcript what the marque is about, its capabilities, and truthful on.

Yet erstwhile it came to the remainder of the site’s AI architecture, Expedia.com scored conscionable 33.3% wide – 1 of the lowest successful the full cohort. No JSON-LD system data, nary sitemap declaration, and only a 4th of the contented is delivered server-side. Publishing specified a well-written llms.txt record while leaving truthful overmuch other unaddressed is simply a small for illustration putting a motion successful your model saying “open for business” but forgetting to unlock the door.

This doesn’t mean specified measures are wasted, conscionable that results are acold from guaranteed. But erstwhile has anything been guaranteed to activity successful SEO? It’s why optimization has ever been astir strengthening arsenic galore signals arsenic possible.

Updating your robots.txt pinch elaborate instructions aliases creating a broad llms.txt aren’t replacements for doing the remainder of the activity to amended your site’s AI architecture.

Block, Allow, Or Do Nothing: The Choice Is Yours

Mentions and citations were ne'er the full story; they’re only the first furniture of existent AI visibility. Your marque could beryllium cited regularly and still suffer customers because of inaccuracies aliases hallucinations, aliases because the AI didn’t person capable assurance successful your accusation to urge your products successful the astir applicable conversations.

Airbnb, eBay, Amazon and others aren’t chasing mentions (in the latter’s case, decidedly not). Each 1 has made deliberate, circumstantial choices astir which AI systems tin train connected their content, and which tin enactment connected a customer’s behalf successful the moment.

The decisions that will specify your brand’s AI beingness won’t each beryllium made by your trading team. Some of them whitethorn person already been made for you, hidden wrong default settings, bequest robots.txt files, and information policies written for a different era.

Whether you’ve planned and optimized for it aliases not, AI is almost surely reference your content. But this is still your content. You tin still exert immoderate power complete really AI accesses, interprets, and interacts pinch your brand.

These are each solvable problems – but only if you determine to lick them.

More Resources: