Why doesn't ChatGPT recommend my company?

1. Blog
2. Discoverability & Content

By Saroj Jha · August 3, 2026 · 8 min read

Ask an AI engine this and you get a visibility checklist: fix your Google Business Profile, get consistent NAP details, collect reviews, earn backlinks, add schema. All true, all worth doing, and none of it explains the actual failure.

The checklist ceiling

out of 100 — agent-readiness score

"Contact us" is a non-answer to a machine.

Ask an AI engine this and you get a visibility checklist: fix your Google Business Profile, get consistent NAP details, collect reviews, earn backlinks, add schema. All true, all worth doing, and none of it explains the actual failure.

You can complete every item on that checklist and still score 52 out of 100 on whether an AI agent can do anything with you.

That is not a rhetorical number. It is what the agent-readiness engine returns when you pass every visibility and markup check and nothing else.

The reason the checklist tops out so low is that roughly half the score has nothing to do with being found. It is about what your business has decided to publish, and what it will let a stranger do without talking to a human. No amount of structured data fixes a pricing page that says “contact us.”

None of this is specific to ChatGPT: the same gap applies to Perplexity, Claude and Google’s AI Overviews. Here is the arithmetic, run on a real audit, with the output shown, using the same reproducible engine-output method as My LTV:CAC Is Below 3:1. Which Number Should I Fix First? The same discipline applied to quarterly reporting metrics is in How do I know if marketing is working?

When an assistant can find you and still leaves you out, the gap is usually what you publish, not your markup. On AAJ’s agent-readiness audit, a site that passes every visibility check and nothing else scores 52 of 100. The other 48 points are publishing decisions: a price a machine can read, specs in tables, comparison facts, and one path that completes without a human.

The visibility half of the problem

The engine scores fifteen checks across five dimensions. Eight of them cover findability and machine-readability, which the standard advice already handles:

• Content renders without JavaScript: 14 points
• Structured data on key pages: 8
• llms.txt published: 5. That is this engine’s weight. llms.txt is a 2024 proposal by Jeremy Howard, not a standard^[3]; Google says Google Search doesn’t use such files^[4], and AAJ’s SEO & GEO Readiness Scorer, a different rubric, gives it 2 of 100
• Clean heading hierarchy: 3
• Organization schema, consistent identity: 6
• AI crawlers permitted in robots.txt: 6
• No blanket bot-blocking at the edge: 4
• Key actions aren’t CAPTCHA-walled: 6

That is 52 points. Every one of them is on the usual “get recommended by AI” checklists, and they are all worth doing. The crawler check is the one most often half done, because each assistant reads pages for its answers with a different crawler from the one that trains its models. Blocking GPTBot keeps your pages out of OpenAI’s training; ChatGPT search reads them through OAI-SearchBot, and blocking that one keeps you out of its answers^[5],[6].

The other 48 points are different in kind:

• Pricing published and parseable: 12
• Self-serve path exists: 8
• Specs and features in structured form: 7
• Comparison facts available: 6
• Public API or documented integration: 6
• Claims sourced and verifiable: 5
• Facts consistent across surfaces: 4

None of those are visibility. Publishing your prices is a commercial decision. Offering a route that completes without a sales call is a go-to-market decision. Documenting an API is a roadmap decision. No amount of directory listings or review volume moves any of them.

What a perfect visibility score actually looks like

Pass all eight findability checks, fail everything else:

Two dimensions at full marks. The verdict is still “weak”, and the engine’s one-line summary is the whole problem: an agent can read the site but can’t reliably evaluate or transact.

Being found is the part everyone optimizes. It is also the part that no longer differentiates anything, because it is the part that is easy.

The single most expensive line item is your pricing page

The largest deduction on that site is not schema, or reviews, or backlinks. It is twelve points for pricing a machine cannot parse. The engine’s reasoning is blunt:

An agent comparing options cannot include you. ‘Contact us’ is a non-answer to a machine — it will return your competitor’s number instead.

That is worth sitting with, because it explains the original question better than any visibility answer does. An agent asked to shortlist three vendors does not pause at a contact form. It does not email you. It returns the two vendors whose numbers it could read, and you are not rejected or ranked lower. You are simply absent.

Publishing prices moves the same site from 52 to . One change, no code, worth more than structured data and a clean heading hierarchy combined.

AAJ’s own site scores 75

I ran this on aajconsult.com. Prices are published, specs are in tables, facts are consistent across pages. It scored 75 of 100, a “partial” verdict, with four gaps left: a self-serve path (8 points), comparison facts (6), a public API (6) and sourced claims (5).

Action surface: 6 out of 20. That is the number that matters, and it is the worst score on AAJ’s site.

Every one of the four remaining gaps is a business-model decision. A consultancy selling sprints has no self-serve path because the product is a person’s attention. Nor is there an API through which an agent could buy one. Those are defensible choices, but they are choices, and the engine correctly refuses to let markup compensate for them.

The engine’s closing summary, set beside the visibility-only site, is the part worth printing out:

Three out of four. The one that fails is the one where revenue happens.

The comparison gets built whether you supply the facts or not

The six-point deduction for missing comparison data carries the sharpest line in the rubric:

Agents build comparisons whether or not you supply the facts. Absent yours, they use a competitor’s framing.

This inverts the usual instinct. The conventional reason not to publish comparison pages is that you would rather not draw attention to alternatives. In an agent-mediated market that stops working, because the comparison happens regardless, assembled from whatever the agent can find, which will be your competitor’s page about you, not your page about you.

The same logic applies to unsourced claims. An agent that cannot verify “3.2× pipeline growth” does not simply believe it less. It discounts it against a source it trusts more, and that source is rarely you. Research on AI answers points the same way: in the GEO study, rewriting a source to add quotations, statistics or citations gave it 27% to 41% more of GPT-3.5’s answers, while keyword stuffing gave it 8% less^[2]. This is also why review volume and backlinks help less than the checklist implies: they establish that you exist, not what you cost or what you do.

Run it on your own site

Two commands, no account. The installer asks which coding agent to install for; the paths below assume its shared .agents/skills/ folder:

npx skills add sarojkjha/aaj-marketing-skills --skill agent-readiness-audit
node .agents/skills/agent-readiness-audit/resources/agent-readiness.js --help

Then add one flag for each check you pass. To reproduce the 52-point visibility-only score above:

node .agents/skills/agent-readiness-audit/resources/agent-readiness.js \
  --ssr --schema --llms-txt --clean-headings \
  --org-schema --allows-ai-crawlers --no-agent-blocking --no-captcha

Add --json for machine-readable output, or --all to see what a perfect score looks like.

One caveat the engine states about itself: this is a self-assessment.

Do this

Verify each answer against the live site: view-source for rendering, robots.txt for access, and an actual agent attempt for the action surface. To measure the reach, read and understand layers on a live URL instead of self-reporting them, use the SEO & GEO Readiness Scorer.

Free tool

Run the fifteen checks in your browser

The free Agent Buyability Check scores the same fifteen checks across five dimensions in the browser, with nothing to install, and lists what to fix first.

What to do with the result

If you scored in the fifties with good visibility, you do not have a discoverability problem. You have a publishing problem, and the order is:

1. Publish prices. Twelve points, and the only gap that removes you from comparisons entirely rather than ranking you lower within them. Put them in the page text with the currency and unit. If you also mark them up with schema.org’s Offer type^[10], Google’s guidelines require the marked-up price to be visible on the page^[11].
2. Structure your specs. Seven points. Tables with consistent labels. Say what it does, not how transformative it is.
3. Open one self-serve path. Eight points. A trial, a purchase, or a booking that completes without a human.
4. Publish your own comparison facts. Six points, including what you don’t do.

The first two are copy changes. The third is a real decision, and for some businesses the right answer is to accept the deduction knowingly, which is a different thing from not knowing it exists. AAJ’s published rates, including the AI visibility work, are on the pricing page: the first fix in this list, applied to AAJ itself.

If you want me to look at your own funnel, positioning and channels, my Growth Audit ends with three prioritized fixes and a 30-day action plan.