llms.txt, schema, and freshness: the 7 fixes that make AI cite you
July 5, 2026 · 10 min · Spike research
Every Spike Fix Pack contains seven artifact types. That number isn't arbitrary — it's the complete set of on-site changes that measurably move how AI engines see a B2B product. Here's each one: what it is, why engines care, and how to ship it.
1. JSON-LD structured data
Schema markup is the difference between an engine inferring what you are and knowing it. Four types matter for SaaS:
Organization— who you are, canonical name and URLSoftwareApplication— what the product is and doesOffer— that pricing exists and whereFAQPage— quotable question-answer pairs
Engines quote schema facts with far higher fidelity than they paraphrase prose.
A SoftwareApplication node with a crisp description is the closest thing to
dictating your own one-liner in AI answers.
Ship it: paste the JSON-LD blocks into your <head>. Validate with
Google's Rich Results test. Twenty minutes.
2. llms.txt
llms.txt is a markdown file at your web root written for AI crawlers: what
you are, who you're for, where the key pages live. It emerged in 2024 and by
now every major crawler requests it. Without one, models reconstruct your
product from navigation links and hero copy — badly.
Ship it: one page, four sections (what, who for, key pages, full detail
link). Add llms-full.txt with long-form product context for models that
want depth.
3. robots.txt that lets AI in
A remarkable share of B2B sites still block GPTBot, CCBot, or
PerplexityBot — usually a 2023-era decision nobody revisited. Blocked
crawlers can't read you; engines that can't read you recommend someone else.
This is a genuine choice (there are reasons to keep AI out), but it should be
a current choice, made knowing the visibility cost.
Ship it: explicitly Allow: / the ten major AI user-agents. Two minutes,
and the single highest-leverage line in this list for blocked sites.
4. An FAQ page that answers buyer questions
Not your support FAQ. A page that answers the questions buyers type into
ChatGPT: best <category> for <use case>, how do I solve <pain>. Written
in short declarative sentences with a concrete fact each — the format
retrieval engines lift verbatim.
Ship it: ten questions, 2–4 sentences each, FAQPage schema attached,
linked from your main nav so crawlers treat it as canonical.
5. An honest comparison page
<You> vs <market leader> is the highest-intent prompt family in existence,
and if you don't have the page, the answer comes entirely from the leader's
marketing and third-party listicles. The counterintuitive part: the page
must concede real trade-offs. One-sided comparisons get ignored; honest
ones get cited.
Ship it: TL;DR with "choose them if…" included, a factual feature table, linked sources for every claim.
6. Freshness signals
Engines discount stale pages — an unchanged pricing page from 2024 reads as a
possibly-dead product. Three signals fix this: dateModified in schema, a
visible "Last updated" line, and a changelog section.
Ship it: add all three, then keep them honest — update on real changes. Fake freshness is worse than none.
7. Meta titles and descriptions that name your category
Your meta description is often the snippet a retrieval engine reads before
deciding to click through. "We help teams move faster" tells a model nothing.
<Brand> is <category> for <ICP> — <one concrete claim> tells it everything.
Ship it: rewrite your three weakest pages first. Title ≤60 characters naming category + brand; description 140–160 characters with one quotable claim.
The week-long plan
| Day | Fix | Effort | | --- | --- | --- | | Mon | robots.txt + llms.txt | 1 hour | | Tue | JSON-LD on home + pricing | 2 hours | | Wed | FAQ page | Half day | | Thu | Comparison page | Half day | | Fri | Freshness + meta rewrites | 2 hours |
Then re-scan weekly. Retrieval-based engines typically reflect on-site changes within two to four weeks of re-crawling; memory-based answers follow slower but compound.
Spike generates all seven artifacts pre-filled from your actual site — run the free scan and download your Fix Pack.