A lot of what is currently sold as AI optimisation is unmeasured. The tactics get described, the invoice gets sent, and nobody establishes whether the brand was cited before or after. Start with measurement instead — it is the only part of this that is not speculation.
First, check whether you actually have a problem
Plenty of brands assume AI answers are eating their traffic when the real cause is ordinary. The check takes five minutes in Search Console. Compare impressions against clicks, by query, across the period your traffic fell:
- Impressions held, clicks fell — you still rank; something above the result is absorbing the click. That is consistent with an AI answer, though a new ad block or SERP feature does the same thing.
- Impressions and clicks both fell — you lost rankings. This is not an AI problem, and the traffic drop diagnostic is where to start.
Only the first pattern justifies spending anything here.
How to measure citation rate
Any single AI answer is noise. Ask the same question twice and you get different text, sometimes different sources. Measuring one answer tells you nothing.
What works is a fixed prompt set measured as a rate:
- Build 30–50 prompts from your real commercial queries — the questions a buyer actually asks, not generic industry phrasing.
- Run the set across the engines your audience uses, and record whether your brand is named and whether your domain is cited.
- Repeat on a schedule. The number that matters is the percentage of prompts where you appear, tracked over time.
- Record who is cited when you are not. That list is the most actionable output.
This is unglamorous and it is the only way to know whether anything you do afterwards worked.
What carries over from ordinary SEO
Most of it, which is worth saying plainly rather than dressing up as a new discipline.
Being crawlable and indexable. Answer engines draw heavily on the existing index. If a page cannot be crawled, it cannot be cited.
Clear, extractable answers. Content that states its conclusion directly, near a heading that matches the question, is easier to quote. This is the same structure that wins featured snippets.
Entity clarity. This is the piece most sites are missing. Answer engines need to resolve “your brand” to a specific entity. Organization markup, Person markup for authors, and sameAs links to your official profiles are how you make that unambiguous. Our guide to which schema types are worth implementing covers the markup itself.
Corroboration across sources. Being described consistently on sites other than your own — directories, publications, profiles — makes a model more confident in asserting something about you. This is conventional digital PR with a different payoff.
Author credibility. Named authors with real, linkable histories, rather than “Admin” or a house byline.
What does not carry over
Keyword density and exact-match phrasing. Models work on meaning. Repeating a phrase does not increase the chance of being quoted.
Ranking position as a proxy. Position one is not required for citation, and holding it does not guarantee it. They are correlated, not equivalent.
Volume. Publishing more does not help if none of it is quotable. A single genuinely authoritative page outperforms twenty thin ones here even more sharply than in ordinary search.
The honest state of it
There is no submission process. There is no vendor with privileged access. Nobody can guarantee a citation, and a supplier claiming otherwise is telling you something useful about themselves.
What is reliably true: entity clarity, structured data, credible authorship and genuinely useful content improve your odds, and all four are worth doing regardless of what happens to answer engines next year. Treat anything more specific as provisional and do not rebuild your site around it.
If you would rather have the baseline measured properly — a real prompt set, a citation rate, and a list of who is cited in your place — that is our AI overview visibility audit.