How to Show Up in AI Answers, and What Nobody Can Promise

The question turns up constantly, in some form: what do I actually have to do to get my company named when a buyer asks ChatGPT for a recommendation? It deserves a straight answer, and the straight answer comes in two halves. There is a set of things you can genuinely do, and they are unglamorous. There is another set that is simply not available to you, whatever anyone tries to sell you.

Most writing on this subject keeps the line between those halves deliberately blurry, because the blur is where the selling happens. So let us draw it.

What actually depends on you

Being readable by a machine, not just by a browser

This is the floor, not the strategy. If a crawler cannot fetch your page, nothing downstream matters. Open yourdomain.com/robots.txt and see whether GPTBot, ClaudeBot, PerplexityBot or Google-Extended sit under a Disallow: /. Then confirm your pages return real text to a plain fetch, rather than an empty shell that only fills in once a browser has run your JavaScript. Clearing this earns you nothing by itself. Failing it costs you everything, which is a bad trade to make by accident.

Saying what you are in sentences that survive being lifted out of context

Models reuse language that explains itself. A sentence that only makes sense while surrounded by your visual identity, your navigation and a carefully staged hero image is a sentence a model can do nothing with.

Here is a check that takes a minute. Copy the main descriptive sentence from your homepage. Paste it into a blank document, alone, no logo, no context. Read it back. Does it tell a stranger what you sell, who buys it, and how it differs from the alternative? Or does it say that you empower organisations to unlock their potential through innovative solutions? The second kind is not merely weak marketing copy. It is copy that is structurally unusable by a system whose entire job is to compress the web into a paragraph.

The version that works is boring and specific. This product is a category noun. It is bought by this kind of company. It does this particular thing and deliberately does not do that other thing. Prose like that is easy to quote and hard to get wrong, which is the combination you want when a machine is deciding whether it can safely say your name.

Answering the question in words, on the page

A model cannot cite what you never wrote down. If your pricing lives behind a demo request, you are not in the answer when someone asks what things cost in your category. If your specifications live in a downloadable datasheet, they are far harder to reuse than the same table in HTML. If you refuse on principle to say how you differ from the obvious alternative, the comparison still gets written, by someone else, their way.

Every question your sales team answers on calls is now being asked to a machine instead. The ones you never committed to text are the ones where you are absent by default.

Keeping your identity consistent wherever it appears

Structured data, product and organisation markup, and the plain facts of your name, category and description should agree across your site, your documentation, and the profile pages that already exist about you elsewhere. Not because a hidden ranking factor rewards schema. Because when the same company is described three different ways in three places, the web is describing three vague entities instead of one clear one, and clarity is the only currency here.

The uncomfortable part: most of the work is not on your site

A generative model did not learn about your industry by reading your homepage. It learned by reading the whole web discussing your industry. Third party documentation. Forums where practitioners complain about tools. Review platforms. Comparison articles by people with no relationship to you. Question and answer threads. Job postings, which quietly reveal what companies actually use.

Your website is one source among thousands, and a source a model has good reason to discount, because every company on earth claims to be the leading provider of whatever it provides. What other people say about you carries weight your own claims cannot buy.

Which means the highest leverage work sits off your property, and it is slow. Publish documentation clear enough that others can quote it without rewriting it. Answer real questions where your buyers actually ask them, under your own name, without pretending to be a neutral bystander. Make sure the profiles that already exist about you are accurate rather than abandoned. Give the reviewers who cover your category a factual page they can copy from instead of having to interpret you.

What does not work, plainly: manufactured reviews, forums seeded with fake accounts, a network of thin sites that all praise you. It breaks the rules of the platforms you would be doing it on, and it is structurally fragile, because it produces exactly the repetitive, low-substance text the web is already drowning in.

What does not depend on you at all

Now the other half of the honest answer.

There is no submission form. Search engines gave you a console, a sitemap, an index you could inspect and a button to request a recrawl. Generative engines give you none of that. There is no endpoint where you register a brand, no ranking factor to tune, no dial. A model's knowledge is shaped by a corpus assembled by someone else, at a moment you did not choose.

You do not choose which names appear. Whether a model names you, where it places you in a list, and how it characterises you, are outcomes you influence only indirectly, through what exists in text about you. Influence is not control, and the gap between the two is where every guaranteed placement pitch makes its living.

The same question does not produce the same answer twice. These systems are probabilistic. Ask a model to recommend vendors, ask again in a fresh conversation, and the list can change. That is not a defect to be optimised away. It means a single check tells you close to nothing, and any conclusion drawn from one conversation is an anecdote wearing the clothes of evidence.

Timing is not yours either. A model can know about you by two routes: what it absorbed during training, and what it fetches live while answering. Something you publish today can be retrieved today by a system that browses, and might never enter another model's trained knowledge.

So the honest expectation for anything you change is not a result by a certain date. It is that you have removed a reason for the machine to ignore you. Removing an obstacle is not the same thing as causing an outcome, and anyone who says otherwise is describing a mechanism that does not exist.

Which leaves measurement

If you cannot guarantee the outcome, the only rational posture is to watch it. Not once, and not only in your home language: across the platforms your buyers use, across the markets you sell into, often enough that patterns separate themselves from noise.

That is what PSentry is built for. It runs a prompt set across ChatGPT, Claude, Gemini and Perplexity, in each language and market you operate in, and reports where you are named, where you are cited, and which competitors get recommended in your place. Consistent with everything said above, the tool does not add a dial the models themselves refuse to give you: it watches what ChatGPT, Claude, Gemini and Perplexity already say and reports it back, nothing more. It does not promise an outcome, because no outcome here is anyone's to promise. And because refreshing that watch every hour would only feed you more noise to misread, the scans run on their own rhythm, twice inside a month, a reading rather than a running commentary.

The brands that handle this era well will not be the ones who found the trick. They will be the ones willing to learn what the machines say about them, accept how little of it is under their control, and keep working on the part that is.

Frequently Asked Questions

If I fix my robots.txt today, when will I appear?

Possibly never, and certainly not on a timetable you can plan around. Allowing a crawler makes you readable going forward. It does not retroactively insert you into knowledge a model was already trained on.

Should I write content specifically aimed at AI models?

Write for a reader who does not yet know your category exists, and state your facts plainly rather than persuasively. That happens to be the format machines reuse best. Text written to game a model tends to read like text written to game a model, which is also how it reads to your buyers.

Does publishing more articles make me more visible?

Not by volume. Nothing counts how many posts you published. What changes a model's picture of you is whether clear, specific, quotable information about your product exists in text, and whether other people repeat it. One reference page that others actually cite is worth more than a content calendar nobody reads.

Is any of this different from good SEO?

The foundations overlap heavily: crawlability, clarity, being talked about elsewhere. The divergence is what you compete for. Search is a competition for placement in a list. A generative answer is a competition for inclusion in a sentence with room for a few names, and there is no second page to climb from.

Can an agency guarantee my brand appears in AI answers?

No, and the promise itself is a useful filter. There is no submission mechanism and no ranking dial. What an agency can legitimately do is improve the raw material the models read about you. What none of them can do is decide the output.