Measure, fix, corroborate, then measure again.
Improving AI search visibility comes down to four stages in a fixed order: measure your citation rate across every answer engine, make your facts machine-readable and consistent, build independent sources that corroborate them, then re-run the identical prompt set every month to prove whether it worked.
Measured, not guessed at.
Every stage produces a number you can check. Nothing here depends on you trusting our judgement over your own reporting.
We test before we talk
We run a dozen real customer questions against ChatGPT, Gemini, Perplexity, and Google AI Overviews — for your business and for two competitors you name. You get the scorecard either way. You see the gap before you spend anything on fixing it.
Make the facts extractable
Schema markup, name-address-phone consistency across directories, a complete Google Business Profile, and your highest-value pages restructured to answer questions in the first line. This is the work that makes a model able to quote you.
Give them sources to trust
Reviews, directory presence, local press, and content worth citing. Answer engines weight what independent sources say about you far more heavily than what you say about yourself, so this is where durable gains come from.
Same prompts, every month
We re-run the identical prompt set each month and report three numbers in order: citation rate, AI-referral and branded search activity, then customers who told you they found you through AI. Always against the same named competitors.
Three numbers. No vanity metrics.
Impressions, reach, and “AI mentions” are easy to inflate. These three are not, which is why they are the only ones on the report.
Citation rate
How many of your tracked prompts name you, broken down per engine. This is the first thing to move, usually within eight to twelve weeks, and it moves before anything downstream does.
AI referrals & branded search
Visits arriving from AI tools, plus people searching your business name directly. This confirms the citations are reaching real customers rather than just existing in a test.
Attributed new customers
Customers who told you they found you through an AI assistant. It lags the other two by design — and it is the only number that decides whether this was worth paying for.
The tactics we refuse to use.
An answer engine cites you because it trusts your information. Every shortcut below works against that, and most of them carry legal exposure on top. This list is part of the engagement, not marketing.
No fabricated reviews or ratings
We do not write, buy, incentivise, or filter reviews. Review generation means asking real customers at the right moment — nothing else.
No invented locations or credentials
We will not create addresses you don't operate from, service areas you don't serve, or claims about licensing, awards, or history that we can't corroborate.
No competitor-adjacent domains
No typosquatting, no lookalike domains, no pages built to intercept a competitor's name. It is trademark exposure and it is not our business model.
No seeding public repositories
We do not plant packages or code in public repositories to influence what AI coding tools recommend. That is supply-chain manipulation regardless of intent.
No scraping in breach of terms
Prospect research uses publicly available business information within platform terms. We do not run automated harvesting that violates them.
No guaranteed AI placement
We will not promise you a spot in an AI answer, or forecast revenue from in-chat checkout. Both are outside our control and anyone claiming otherwise is selling a guess.
We ran this on ourselves first.
Before selling AI visibility work, we applied the whole checklist to this website.
Structured data on every page. Content written to answer a question in its first sentence. HTML rendered on the server, so a crawler that never runs JavaScript still sees everything. Explicit access granted to GPTBot, PerplexityBot, and ClaudeBot in our robots.txt. A real dateModified on every page, matching the review date printed at the bottom of it.
The page you are reading is the first audit we ran. Every item on the foundation checklist we would hand a client is one we have already done here — you can verify all of it by viewing source.
We are not publishing client citation numbers, because we do not have them yet. Our own before-and-after is being documented now, and pilot case studies follow after that. When there are results worth showing, they will appear here with the prompt set and the dates attached.
The kind of question we test
Not keywords — full questions, in the words your customers use. Localised to your category and metro before we run them.
This is the method. Start with the audit ↗ to see where it would apply to you.
See where you stand before you spend anything.
Tell us your business and the one or two competitors you lose to most. We'll run the prompt set against all four answer engines and send you the scorecard — including the competitor comparison.
- hello@develup.studio
- Remote — serving businesses across the US
- Run and reviewed by a person, not a tool
Reviewed and updated