1. Measure a real baseline
Build a fixed set of prompts that reflect actual buying situations: category discovery, use-case recommendations, comparisons, geographic constraints, integrations, budget and alternatives. Record whether your brand is mentioned, whether your own domain is cited, which competitors appear and which external sources the answer relies on.
A branded prompt such as “What is Acme?” only proves that the system can discuss a brand when explicitly named. The more useful tests are unbranded prompts where a buyer could discover you or a competitor.
2. Fix access before content
Check robots rules, HTTPS, canonical URLs, sitemap discovery and whether important pages are public and render meaningful text. Different providers publish different crawler controls, so verify the search/retrieval bots you actually want to permit rather than treating every AI crawler as the same thing.
3. Make the entity unambiguous
State what the company is, what it sells, who it serves, where it operates and how it differs. Keep those facts consistent across the site and legitimate third-party profiles. Use Organization or relevant schema where it accurately reflects visible content, but do not assume schema alone creates authority.
4. Publish answer-shaped evidence
For every important buyer question, ask whether your site contains a concise, useful passage that could support a good answer. Strong pages define the problem, state the answer early, explain tradeoffs, include concrete facts, and link to supporting evidence. They do not bury the useful sentence beneath generic introductions.
5. Strengthen independent corroboration
Your own site is a claim you make about yourself. Independent sources can corroborate that claim. Relevant editorial coverage, trusted directories, customer evidence, industry references, community discussions and expert citations can all help establish a clearer public footprint when they are genuine.
Do not manufacture reviews, fake press or low-quality link networks. Those create noisy evidence and reputational risk rather than durable authority.
6. Separate mentions from citations
A brand mention tells you that the company entered the answer. A citation tells you which source was used as evidence. A recommendation tells you something stronger: the model placed the brand into a decision context. Measure all three separately so you know what changed.
7. Re-test the same prompts
Keep the prompt set stable enough to compare periods. Record provider, date, market assumptions and whether web search was enabled. Look for directional movement across a group of prompts rather than celebrating one favorable answer.