There is a lot of noise on this subject, most of it selling a technique. So let us be clear about the mechanism first, because it determines what is worth doing.

Google AI Overviews, Google AI Mode, ChatGPT Search and Microsoft Copilot are not a separate index you can submit to. They answer by retrieving from the open web and then summarising what they retrieved. That means eligibility runs on the same foundations as ordinary search, with a few specifics that are easy to get wrong and one or two that are genuinely new.

It also means nobody can promise you a citation. Anyone who does is selling something.

Start with access, because nothing else matters if this is wrong

A retrieval system that cannot fetch your page cannot cite it.

Check that your robots.txt does not block the crawlers you want, that your pages return 200 rather than a challenge or an error, and that nothing in your firewall or bot protection is quietly refusing them. This sounds obvious. It is also the single most common reason a site is invisible to one of these systems while being perfectly visible in normal search.

OAI-SearchBot and GPTBot are not the same thing

This distinction causes more accidental self-harm than any other item on this page.

OAI-SearchBot GPTBot
What it does Retrieves pages for ChatGPT Search results Crawls pages for model training
If you block it You become ineligible to be shown and cited in ChatGPT Search Your content is excluded from training data
Effect on discoverability Direct. You disappear from that surface None

A business that decides it does not want its content used for AI training, and blocks both, has also removed itself from ChatGPT Search without intending to. Those are two separate decisions and should be made separately. The training question is about content licensing. The retrieval question is about whether customers can find you.

One caution on implementation: once you declare a crawler-specific group in robots.txt, that crawler stops reading the general User-agent: * group entirely. Any rule you later add to the general group silently stops applying to it. If a global Allow already covers the crawler, adding a specific group is not a safety measure, it is a future trap.

Be unmistakable about who and where you are

Retrieval systems have to resolve your business to an entity before they will name it in an answer. Ambiguity costs you the mention.

Practically:

  • Use the same business name, address and contact details everywhere. Website, Google Business Profile, Bing Places, directory listings, social profiles.
  • Publish LocalBusiness structured data with the real address and service area, and link it to your Organization entity rather than repeating a separate copy on every page.
  • Include sameAs links to the profiles that genuinely belong to you. Do not invent them.
  • If a detail does not exist yet, leave it out. Missing data is better than inconsistent data, because inconsistency actively reduces confidence in the entity.

A note from our own site: we do not publish a public phone number, so our structured data does not contain one. An invented or inconsistent number would be worse than the gap.

Make the answer readable

This is where a lot of otherwise good sites lose.

Retrieval works on the rendered page. If the sentence that answers the question only exists inside an image, a canvas element, a WebGL scene, a video, or a component that requires JavaScript to produce any text at all, it is not available to be quoted.

Practical version:

  • State the direct answer in plain HTML, near the top, in a sentence that makes sense on its own if lifted out of context.
  • Use real headings, real lists and real tables rather than styled divs.
  • Keep the page structure logical in DOM order, not just visually.
  • If a summary component collapses content, make sure the text is present in the initial markup rather than injected on open.

Decorative animation is fine. Load-bearing animation is not.

Local prominence still does the heavy lifting

For “who does X near me” questions, the answer usually leans on the same local signals that drive the map pack.

  • A complete, accurate and active Google Business Profile.
  • A Bing Places listing, which matters more than Bing’s search share suggests, because Bing grounds Copilot.
  • Genuine reviews, which contribute to prominence.
  • Consistent name, address and phone details across the web.

If you do one thing after reading this and you are a local service business, make it the business profiles.

Give the systems something worth citing

Everything above makes you eligible. It does not make you quotable.

A retrieval system chooses between several pages that could answer the same question. It will tend toward the one that is specific, current, attributable and not obviously recycled. That means:

  • Original information. Data you collected, an outcome you can describe, a method you actually use. Summarising what everyone else already published gives the system no reason to prefer you.
  • Named authorship. A real person, with a real profile, whose expertise is verifiable. Do not manufacture credentials, and do not rotate authors to manufacture the appearance of a team.
  • Dates that mean something. A published date and an honest updated date.
  • Sources. Where you make a factual claim about someone else’s product or policy, link to their documentation rather than to another blog summarising it.

On llms.txt

You will see this recommended. Google has stated it is not used for visibility in Google Search or its generative features, so adding one as a Google tactic is cargo cult work.

If a specific system you care about documents that it consumes the file, implement it for that system and note what you expect to gain. Otherwise it is a file nobody reads.

What you can and cannot measure

Be realistic here, because the reporting is genuinely uneven.

  • Google exposes some generative performance data in Search Console. Use it as the primary source for that surface.
  • Bing Webmaster Tools reports citations and grounding queries for Copilot, which is the most direct AI visibility reporting currently available anywhere.
  • ChatGPT has no publisher console. You can observe referral traffic, test queries manually and track branded search growth, but there is no authoritative count.

Two warnings. AI answers vary between sessions, so a manual test is an observation and not a ranking. And treat third-party “AI visibility scores” as directional at best, because none of them can see the systems’ actual retrieval.

A practical checklist

Work down this in order. The early items gate the later ones.

  1. Confirm your pages return 200 to Googlebot, Bingbot and OAI-SearchBot, with no challenge or block.
  2. Check robots.txt does not accidentally exclude a retrieval crawler, and decide GPTBot separately and deliberately.
  3. Make sure every important page is indexable, canonical and in your sitemap.
  4. Publish accurate Organization and LocalBusiness structured data, linked rather than duplicated.
  5. Make name, address, contact details and profile links consistent everywhere they appear.
  6. Complete and maintain your Google Business Profile and Bing Places listing.
  7. Put the direct answer to each page’s main question in plain HTML near the top.
  8. Replace anything load-bearing that only exists in an image, video, canvas or JavaScript-rendered component.
  9. Add named authorship and honest dates to substantial content.
  10. Publish something only you can publish, then link your sources where you cite others.
  11. Set up measurement in Search Console and Bing Webmaster Tools, and keep a manual query log for ChatGPT.
  12. Re-check after any redesign, replatform or firewall change, because these break access silently.

The honest summary

You cannot buy, configure or guarantee an AI citation. What you can do is remove every reason a system would be uncertain about you, make your answers machine-readable as well as human-readable, and publish something worth quoting. That is a less exciting claim than most of what is written on this subject, and it is the part that actually works.

This page will be updated as these products change, which they do frequently.