Method
Indexed. Cited.
Recommended.
Three stages, run in order, because each depends on the one before it. Below is the whole method, including the parts that argue against things our competitors sell.
The method at a glance
- Stages, run in order
- 3
- Answer engines measured
- 5
- Queries in a golden set
- 40
- Runs per query, per engine
- 3
- Observations per month
- 600
- Claims we will not make
- 4
Stage 01
Indexed
A retrieval system can only cite what it can resolve.
Answer engines parse JSON-LD to confirm entities, dates, authorship, ratings, and relationships before they trust a page enough to name it. Most practice websites offer prose where the machine needs facts, and the machine responds the way it always does with ambiguity: it picks someone else.
Entity architecture
Every provider, procedure, location, and credential modeled as a resolvable entity in JSON-LD, cross-referenced to the external records the engines already trust. Conflicting facts across your site, your directory listings, and public registries get reconciled, because engines resolve conflict by dropping the entity entirely.
Answer-shaped content
Every page opens with the direct answer to the question it exists to serve, in the language a person actually types, before any positioning copy. Pages that state a clear answer near the top are materially more likely to be cited. Welcome paragraphs are a retrieval liability.
Crawler access
GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and Bingbot each verified against the live site, not against a config file. We find these blocked by an inherited robots rule or a WAF default on roughly every other site we audit.
Technical floor
Server-rendered content rather than client-only rendering, valid semantic structure, Core Web Vitals passing on live URLs at launch. None of this is exciting. All of it is disqualifying when absent.
Stage 02
Cited
Legibility makes you eligible. Corroboration gets you named.
Once a page can be read, the question becomes whether anything outside it agrees. Brand search volume and web mentions are the strongest known predictors of citation, ahead of traditional backlinks: brands in the top quartile for mentions see roughly ten times the AI visibility of the bottom quartile. That is an off-site problem, and it does not get solved by publishing more blog posts.
Answer coverage
We hold a live map of the questions your market asks and which ones you have no page for. Coverage expands against the gaps, not against a keyword list bought from a tool.
Corroboration and mentions
Digital PR, directory and registry hygiene, association and credentialing profiles, and the specific citation surfaces each engine over-indexes on. Different engines trust different corners of the web, so this is engine-specific work, not one campaign.
Review corpus health
Ratings are parsed as a trust signal before a practice is named. Volume, recency, and distribution all matter, and a review engine that produces them at a steady cadence beats a burst that reads as manufactured.
Entity maintenance
Providers leave, credentials renew, locations move, insurance networks change. Every one of those is a fact an engine has cached about you. Stale facts contradict the site, and contradiction costs citations.
Stage 03
Recommended
A position you did not measure is a position you cannot defend.
Answer engines are non-deterministic. Ask the same question three times and you can get three different sets of sources. Any report built on a single run is noise presented as progress. We fix the method and let the number move on its own.
Answer Share, monthly
Forty queries agreed with you at kickoff, run three times against five engines, every month, unchanged. Same instrument, same query set, so the number is comparable to itself across a year.
Losses reported alongside wins
You see the runs where a competitor was named instead of you, and who it was. A report that only contains good news is not a report.
Explicit coverage gaps
If an engine was unreachable during a sweep, that shows as a gap rather than a smaller denominator. Quietly shrinking the denominator is the oldest way to manufacture a trend.
Revenue attribution
Call tracking, form attribution, and booking events wired at build time so the monthly conversation is about booked cases, not sessions. Traffic is an input. You cannot deposit it.
What we will not claim
The inconvenient half of the evidence.
A firm that only tells you the findings that support its invoice is not reading the research. It is shopping in it.
llms.txt does not currently move citations
Four independent studies converge on no measurable lift. Microsoft and OpenAI crawlers do fetch it where present, and Google states it does not use it. We ship one because it takes an hour and costs you nothing. Anyone selling it as a strategy is selling you an hour of work at a strategist's rate.
There is no universal optimization
Only about eleven percent of domains cited by ChatGPT are also cited by Perplexity. The engines run on different retrieval stacks and different trust surfaces. A single tactic that wins everywhere would be a remarkable finding, and no one has produced it.
Nobody can guarantee a citation
The systems are non-deterministic and the vendors change retrieval behavior without notice. We guarantee the method, the cadence, and the reporting, and we put a remedy in the contract if the number does not move. That is the honest shape of a commitment here.
Some practices should not hire us
If your buyers do not research before they call, or your case value cannot carry the fee, the arithmetic does not work and we will say so in the audit. We would rather lose the engagement than defend it at month nine.
Where these figures come from
- Analysis of 680M citations across ChatGPT, Google AI Overviews, and Perplexity (domain overlap between engines)
- 2026 study of 34,234 AI responses across platforms (brand citation rate variance)
- Evertune analysis of 200M prompts (citation concentration ceiling)
- Limy, OtterlyAI, ALLMO, and SE Ranking (llms.txt impact studies)
- Service-side documentation from OpenAI, Google, and Microsoft (crawler behavior and directives)
Published third-party research current as of August 2026. Figures on this site are attributed rather than asserted, and we update them when the underlying studies do.
Start here
Find out where you actually stand.
Every engagement opens with the Answer Audit. Forty queries, five engines, three runs each, and a written baseline. If the number is already good, we will tell you that and you will have spent $4,500 to stop worrying.