AI search
Be the source the answer cites.
A buyer asks an assistant instead of a search engine and gets one paragraph with a couple of links. Those links are pages that could be fetched, parsed and quoted. That is a publishing problem before it is a prompt problem.
Crawlable by defaultStructured data per postAn llms.txt index of the archive
Four questions that stopped being searches.
Each of these used to be a keyword research session and a landing page. Now it is a prompt, answered from whatever the assistant can read — which is either your page, a competitor's, or nobody's.
- Who should I hire for this?
- The buyer no longer types a keyword and reads ten blue links. They ask, and get one paragraph with two or three sources. Whether your business can appear in that paragraph depends on whether a page of yours answers the question in a form a machine can lift.
- Which tool actually does this?
- Category and comparison questions are where assistants are relied on most, because the buyer is trying to skip a demo call. A comparison the model can quote — both sides named, the difference stated plainly — is the shape that gets used.
- What does the rule require?
- Regulated questions get answered with a citation to whoever wrote the clearest page about it. Specifics beat prose: a threshold, a deadline, the article of the law, each attributed to the body that made it.
- What does this normally cost?
- Price questions are answered with numbers someone published. State your own price on your own page and it is quotable; leave the market's numbers unstated and the assistant quotes somebody else's estimate instead.
What the crawler has to be able to do with the page.
Four mechanical properties, and not one of them is a writing tip. This is the part that decides whether a page is in the pool an assistant can read from at all; what the page then says is the next page's problem.
- 01
Fetchable
Plain server-rendered HTML, no login wall, and a robots.txt that does not turn the crawler away. A hosted blog answers robots.txt with `User-agent: *` and `Allow: /`, plus its own sitemap line — nothing is disallowed.
- 02
Parseable
A canonical URL, a sitemap entry, BlogPosting structured data with the language and the date, and a FAQPage wherever the article answers questions. Machine-readable facts, not only readable prose.
- 03
Dated
A publish date in the markup, and a last-updated date when a page changes. An assistant choosing between two versions of the same fact uses the one it can place in time; a page with no dates is a page it cannot check.
- 04
Attributable
Published on your domain, dated, and indexed. When a model quotes you it links to you — which is the whole difference between being cited and being paraphrased without credit.
Who comes to read it.
A hosted blog answers robots.txt with an allow-all rule and a line pointing at its own sitemap. As of writing, the crawlers behind the assistants and their user agents are these — the list changes, so check each vendor's documentation rather than our page.
| Model | Crawler | Hosted blog default |
|---|---|---|
| ChatGPT / SearchGPT | OAI-SearchBot, GPTBot | Allowed |
| Claude | ClaudeBot, Claude-User | Allowed |
| Perplexity | PerplexityBot | Allowed |
| Google AI Overviews | Googlebot (Search) | Allowed |
| Gemini training | Google-Extended | Allowed |
The same work, in every market you sell in.
An assistant answers in the language the question was asked in, from sources published in that language. The pages it can use are the ones written in that language: machine-translating an English archive does not put German pages in front of a German question.
Each language is its own editorial line: its own queries, its own dates, its own structured data declaring the language of the article. One site can hold the German, Dutch and English lines without any of them borrowing the others’ answers.
- ENEnglish2 posts a weekplanned / live12 / 7
- DEGerman1 post a weekplanned / live5 / 3
- FRFrench1 post a weekplanned / live4 / 1
Same blog, same design — each language gets its own plan and its own dates.
Two things worth being straight about.
This is a new enough field that most of what is sold as certainty is not, so here is where the line sits.
What we do not claim
Xt4b does not measure whether an assistant cited you, does not predict which engine will, and cannot promise it happens at all. The mechanisms a publisher controls are public and mechanical — fetchable pages, structured data, direct answers, sourced figures, a date, a canonical URL. Everything past that is someone else’s system deciding, and it is the reason the guarantee on this site is about what gets published rather than what it earns.
The rest is churn we do not control: crawler policy, model behaviour and which engines read a file change under you, and the terms of some of these vendors are still being written. A page that is crawlable, dated and specific survives that churn; a trick aimed at this quarter's interface does not.
Structured to be extracted, not skimmed.
Both are content. Only one of them survives being read by something that was never going to scroll.
Content for a results page
- One page per keyword, each competing with the next for the same query
- Figures stated without a source, because a human reader forgives it
- Nothing machine-readable beyond the HTML: no structured data, no index file
- Written for one market, then translated when a new market opens
- Published when the backlog allows, dated whenever the CMS happened to run
Content for a retrieval pipeline
- One query per post, mapped against your sitemap so two pages never chase one question
- Every figure traceable to a named source, or removed before publishing
- BlogPosting and FAQ structured data per post, plus a sitemap entry and a date a parser can read
- A plan per language, written for the queries people type in that language
- A dated cadence, so the archive has a shape a retrieval system can read
The parts of your business no crawler can reach.
A retrieval system quotes what it can fetch. Everything that only exists behind a form, a login, a PDF or a phone call is invisible to it — which is a publishing decision rather than an accident, and usually a deliberate one.
What a plan decides before writing- Pricing that lives in a PDF or behind a demo request rather than on a page
- A comparison table, a changelog or a docs page that answers the question and then cannot be read
- The claims only a conversation can qualify, which a quoted sentence will state flatly
- Which pages you are willing to be reduced to a single paragraph
- Whether training crawlers are welcome, or only the ones that answer questions
Related use cases
- GEO for AI answers
GEO without the theatre: one answer per page, quotable and attributable.
- llms.txt
The summary file for language models, and what ours contains.
- For AI Overviews
Publish the page the block can summarise.
- For multilingual markets
Which market to open next, and what it costs.
Questions about AI search.
Including the one we answer with a no.
Can you guarantee ChatGPT or Claude will cite us?
No, and be careful of anyone who does. What Xt4b controls is what it publishes: pages a crawler can fetch, structured data a parser can read, answers in the first sentence, figures with sources, a date, and a canonical URL on your own domain. Whether a model lifts any of it is the model's decision, made by systems nobody outside those companies can see.
Do you report AI visibility?
No. Xt4b reports what your Search Console reports: clicks, impressions and average position for the posts it publishes. Tools that track mentions inside ChatGPT, Claude or Perplexity are a different product category, and our comparison pages say plainly where those tools are the better pick. We would rather leave a gap in the feature list than put an unverifiable number on your dashboard.
What happens if an assistant quotes us wrongly?
You cannot edit the model, but you can edit the page it read. A wrong answer is usually a correct copy of something you published and have since changed — so the lever is the source: fix the page, keep its date current, and the next fetch carries the correction. Where the answer is right but incomplete, the fix is to state the missing fact on the page that gets quoted, not to argue with the interface.
Does this replace keyword research?
No, it changes what you do with it. The query still comes first: a post is planned around one search, with the intent written down and the sitemap checked so you are not answering a question you already answer. The difference is what the post has to contain once it exists — a direct answer, a structure, a source for every number — so that it is usable both on a results page and in an answer.
Which crawlers can reach the blog?
Every one, by default: a hosted blog serves a robots.txt that allows all agents and disallows nothing at all, plus a line pointing at the blog's own sitemap. That includes the search crawlers (OAI-SearchBot, ClaudeBot, PerplexityBot, Googlebot) and the training ones (GPTBot, Google-Extended). If you would rather opt a training crawler out, ask us — that directive is served by the blog host, so it is not something you can change from your own site's robots.txt.
How long until an assistant picks it up?
Unknown, and it is honest to say so: assistants re-crawl on their own schedules, and they answer from whatever they had when the question was asked. What Xt4b controls is that a page exists, that it is dated, and that the next one arrives on its date — which is the part of this that compounds.
Start with one market
Publish pages a machine can quote.
Connect the domain and pick a language: the first week of entries lands on the calendar, and every post ships with its answer, its sources, its date and its canonical URL intact.
Structured data on every postNo invented AI-visibility promises3 days free, no card