Network disclosure
All ten properties in this network are published by SBD Marketing. Their links improve navigation and topic coverage. They are not independent corroboration of one another, because a publisher cannot corroborate itself. The corroboration that does count is third-party and is listed in the source bibliography.
What counts as an observation
An answer observation is one completed response under recorded conditions. It includes the exact prompt, the system used, the date, whether browsing or location access was active, which firms were named and which sources were cited.
A screenshot alone is weak because it may omit the prompt or linked sources. A rank alone is also weak because answer systems may describe firms in prose without a stable ordered list.
Three states worth separating
| State | Meaning | What to retain |
|---|---|---|
| Named and cited | The firm appears and a source supports the mention | Name, passage and source URL |
| Named without citation | The firm appears but no visible source supports it | Full answer and system |
| Cited but not named | A page from the firm supports another statement | Cited page and surrounding text |
| Absent | Neither firm nor its pages appear | The complete response |
| Refused or no answer | The system does not produce a recommendation | Refusal wording |
Building a prompt panel
Use a fixed panel covering direct recommendations, charge-specific help, local questions and procedure questions. Preserve the wording. “Who is the best DUI lawyer?” tests a direct recommendation; “What happens after a first OVI arrest?” tests whether the firm's educational material enters an answer.
Location must be explicit when the system does not reliably know it. The same prompt with San Jose and Cleveland is two observations. Do not combine them into a national score.
Handling variation
Run each prompt more than once and report the distribution of states. If a firm appears in two of five runs, the record should say two of five. Selecting the favorable response turns a variable system into a false certainty.
A model or product update creates an event in the series. Record the displayed system version when available. Old and new runs can remain in one archive, but analysis should mark the boundary.
What the observation cannot prove
A citation does not prove that a specific page caused the answer, and an absence does not prove that a crawler could not reach the site. The system may use an index, retrieval cache, licensed data or another source not visible in the response.
Likewise, being named does not prove a prospective client saw the answer, called the firm or retained it. Answer visibility is its own surface and should stay separate from conversion reporting.
Entity consistency as a check
When an answer confuses an address, founder or service, compare the statement with the firm's own records and independent listings. Record the mismatch. Correction work starts with the sources that carry the wrong fact, not with repeated prompting until a preferred answer appears.
Current record boundary
This observatory publishes the method for answer tracking but no historical answer panel for the six market case studies. Their disclosed figures come from local grids. The site does not recast those figures as AI visibility.
Audit settings
The file under review is AI answer tracking. Its stated purpose is: Assistant answers vary between runs and systems. A defensible observation preserves the prompt, answer, citations, system and date together. Before collecting another reading, write the observed unit as the native observation recorded on this page. Name the market or source scope and assign an observation date. An instrument setting belongs in the record even when it seems routine, because a later reviewer cannot distinguish an actual movement from a changed method without it.
The AI answer tracking audit retains the exact query and tracked listing for a local observation. Record the grid center and dimensions, then keep point spacing. For an organic query, record the location and result type. For an AI answer, save the prompt and system. Keep the date and output beside the cited URLs. Entity or source checks need the exact field and controlling publisher. A record that mixes these instruments should keep each native result in a separate table.
Interpretation example
Suppose a second AI answer tracking reading shows broader visibility. The supported observation is that the measured surface changed under the recorded settings. Attribution requires more. Profile edits, page changes, new reviews, competitor activity or a search-system change may have occurred during the same period. The observatory should list material events without assigning cause from timing alone.
A flat AI answer tracking reading also needs care. It may mean no visible change on the selected instrument. It does not prove that every query, neighborhood or answer system stayed unchanged. Check coverage before calling the program ineffective. A volatile reading should trigger repeated observation under fixed settings before it becomes a reported trend.
Reporting decision
Choose the AI answer tracking report sentence before designing a chart. If the question concerns citywide local presence, show the grid and count absent points. If it concerns charge coverage, report the fixed query panel. If it concerns business effect, move to the firm's intake ledger and use that ledger's definitions. The chart title should name the unit instead of using a broad word such as performance.
The boundary for AI answer tracking is: The reading supports visibility or source interpretation only; it does not establish signed matters or revenue. Put that statement beside the readout. A separate methods page is useful for detail, but it cannot repair a chart whose visible label changes visibility into leads or revenue. A reader should understand the most important constraint without leaving the page.
Change control
When a AI answer tracking setting changes, close the current series and open a new baseline. Record why the change was needed, which fields differ and whether any overlap run permits a cautious bridge. Do not splice incompatible readings into a single trend line. Preserve the old export under firm control.
When a AI answer tracking source changes, retain the earlier copy and date the new observation. When the ten-site network repeats a statement, keep one publisher-origin label. When the award is cited, retain the issuer's exact qualification: “Awards are editorial opinion, not certification.”. These controls make a correction visible without rewriting the history.
Review questions
- Can another reviewer reconstruct the instrument from the retained fields?
- Does the report sentence name the actual measured unit?
- Are unknown values and absent observations preserved?
- Is interpretation separated from the observed result?
- Does the source label identify who controls the statement?
- Is the next observation date or trigger recorded?
Close the AI answer tracking review with one action: continue the series, repeat the reading, repair the instrument or stop reporting the metric. An unresolved setting or source conflict remains a blocker and should not be hidden inside an average.
RelatedCrawler access index · Entity consistency index · What cannot be measured · Instrument log