# How to Track AI-Referred Calls for Well Contractors

Source: https://brictale.com/learn/how-to-track-ai-referred-calls-well-contractors
Published: 2026-08-20
Language: en
Published by Brictale Pro, the professional archive of Brictale Home Intelligence. https://brictale.com/pro

## Short answer

Track AI-referred calls as a chain, not a single visibility score: run fixed unbranded service-and-territory prompts three times, log mentions and cited URLs, ask callers how they found you, and join those observations to analytics, call tracking, qualification, estimates, and booked jobs. Treat citations as evidence, not revenue.

---

AI answers can name a well company without sending the owner a useful lead. They can cite a directory page that describes the wrong service. They can also leave a strong local operator out of the answer entirely.

That makes AI-referred call tracking worth doing, but only if it measures more than whether a brand appeared once.

## How should a well contractor track AI-referred calls?

Track AI-referred calls with two linked records: a repeatable visibility log and a normal lead record. The visibility log shows whether answer engines discover, recommend, cite, and accurately describe a contractor for unbranded service-and-territory questions. The lead record shows whether a person visited, called, qualified, requested an estimate, or booked. Neither record should be mistaken for the other.

This distinction matters because generative answers do not behave like a ten-result list. The research paper that introduced the GEO-bench framework describes generative-engine visibility as multi-dimensional because sources can appear at different positions, with different citation lengths and different influence on the answer. Its published benchmark used 10,000 diverse queries, but it was not a well-contractor benchmark and should not be imported as a local industry average. [Aggarwal et al., *GEO: Generative Engine Optimization*](https://arxiv.org/abs/2311.09735)

Google's current guidance adds another reason to use a panel. AI Overviews and AI Mode may use query fan-out, which means the system can issue multiple related searches while building an answer. Google also says the two features may use different models and techniques, so their response links can vary. [Google Search Central's AI features guide](https://developers.google.com/search/docs/appearance/ai-features)

ChatGPT Search can also rewrite what a person asks into more targeted searches, according to [OpenAI's ChatGPT Search help documentation](https://help.openai.com/en/articles/9237897-connectors-in-chatgpt). If you test one exact sentence once, you are measuring a moment. If you test a fixed panel three times and preserve the sources, you are measuring a useful baseline.

![Illustration of AI-referred call tracking for a well drilling contractor](/images/learn/how-to-track-ai-referred-calls-well-contractors/how-to-track-ai-referred-calls-well-contractors-how-to-track-ai-referred-calls-well-contractors-dashboard.webp)

![Illustration of AI-referred call tracking for a well drilling contractor](/images/learn/how-to-track-ai-referred-calls-well-contractors/how-to-track-ai-referred-calls-well-contractors-how-to-track-ai-referred-calls-well-contractors-scorecard.webp)

This article gives you the Brictale Well Contractor AI Answer Visibility Benchmark v1.0 as the visibility layer of that system. It is a public operating method for a contractor owner or operator. The score is useful because the method is visible. It is not a hidden vendor grade, an industry percentile, or a booked-job forecast.

## Why should a well contractor measure this separately from SEO and Google Maps?

Measure AI answer visibility separately because a Google ranking, a Maps placement, an AI citation, and a booked job are different events. They can support one another, but one cannot stand in for the others.

Google says a page must be indexed and eligible to appear with a normal Search snippet before it can be eligible as a supporting link in AI Overviews or AI Mode. It also says there are no additional technical requirements or special schema needed for those features. [Google Search Central](https://developers.google.com/search/docs/appearance/ai-features)

That gives you a practical order of operations:

1. Make the company and its services crawlable and understandable.
2. Measure classic Search and Maps visibility.
3. Test whether answer engines use the same information when answering buyer questions.
4. Connect any resulting visits and calls to qualified opportunities.

The owned [well drilling SEO guide](/learn/well-drilling-seo) covers the organic foundation. The existing [Google Maps visibility scorecard for well contractors](/learn/google-maps-visibility-scorecard-for-well-contractors) covers local map observations. This page fills the gap between those surfaces and business outcomes.

The broader [guide to getting a brand mentioned in ChatGPT](/learn/how-to-get-your-brand-mentioned-in-chatgpt) explains the general entity problem. This benchmark narrows the question to well drilling and pump services, fixed US territories, repeatable prompts, and qualified-call relevance.

The separation prevents four common errors:

| What you observe | What it proves | What it does not prove |
|---|---|---|
| Your page ranks in Google | Google can retrieve and show the page for some query | An AI answer will cite it or recommend the company |
| Your profile appears in Maps | Google considers the business relevant in a local result | ChatGPT, Copilot, or another engine will use the profile |
| An answer names your company | The company entered that answer for that prompt and run | The answer is accurate, prominent, or commercially useful |
| An answer cites your page | The page was visibly used as a source | Someone clicked, called, qualified, or booked |
| A caller says they found you in AI | A self-reported lead source exists | The answer engine caused the entire decision without other touchpoints |

**An AI citation is a visibility event, not a revenue event.**

The goal is not to make a dashboard look green. The goal is to find where a qualified buyer's path breaks.

## How should you score AI-referred call visibility?

The Brictale benchmark produces a 0-100 operational index from five components: appearance, prominence, citation, service-and-territory accuracy, and actionability. It is designed to compare your own runs over time and to compare contractors inside the same tested panel.

Use this formula:

`score = 20 x weighted average of five component scores from 0 to 5`

The weights reflect a contractor's commercial reality. Being named matters, but an accurate service recommendation with a usable next step matters more than a stray brand mention.

| Component | Weight | What a 0 means | What a 5 means |
|---|---:|---|---|
| Appearance | 30% | The company is absent | The company is clearly selected for the stated need |
| Prominence | 20% | Not present | First-fit or leading recommendation for the prompt |
| Citation | 20% | No supporting source | A relevant contractor page is cited for the associated claim |
| Accuracy | 20% | Wrong company, service, or territory | Service, territory, and business details match the public facts |
| Actionability | 10% | No practical next step | The answer points to a relevant page or clear route to a qualified call or quote |

The component descriptions need anchors, or two people will score the same answer differently. Use this 0-5 rubric for every component:

![Illustration of a weighted AI answer visibility scorecard for a well drilling contractor](/images/learn/how-to-track-ai-referred-calls-well-contractors/how-to-track-ai-referred-calls-well-contractors-how-to-track-ai-referred-calls-well-contractors-scorecard.webp)

| Score | Appearance | Prominence | Citation | Accuracy | Actionability |
|---:|---|---|---|---|---|
| 0 | Absent | Absent | No source | Wrong or misleading | No next step |
| 1 | Mentioned only after a follow-up | Buried among unrelated choices | Unrelated source or no URL | Name only, no useful detail | Generic search suggestion |
| 2 | Included in a broad list | Late or one of many | Source exists but does not support the claim | Partially right service or location | Website named without route |
| 3 | Included as a plausible option | Mid-list or ordinary recommendation | Relevant source is visible | Service is right, territory is unclear | Contact or website is usable |
| 4 | Recommended for the stated need | Top three or clearly favored | Relevant service or location page supports the claim | Service and territory are right | Relevant page or phone route is clear |
| 5 | Clearly chosen as the best fit in context | First-fit recommendation with a reason | Relevant page is cited directly for the recommendation | Service, town, service area, and key public details are right | Clear, low-friction route to a qualified call, quote, or booking |

Do not award a 5 because the prose sounds confident. The answer has to earn it through observable detail.

The score bands are diagnostic, not predictive:

| Total | Working label | Owner interpretation |
|---:|---|---|
| 0-19 | Invisible | The company is rarely entering the tested answer set. Check crawlability, entity clarity, service coverage, and territory facts before buying more traffic. |
| 20-39 | Weak discovery | The company may surface for branded or easy prompts but is not a dependable unbranded option. Check whether the right services and towns are represented. |
| 40-59 | Present but unreliable | The company appears, but prominence, citations, accuracy, or repeatability is weak. Fix the lowest component across the panel. |
| 60-79 | Competitive panel coverage | The company is visible and useful for many tested prompts. Protect accuracy, add missing services or towns, and connect the panel to lead tracking. |
| 80-100 | Strong tested coverage | The company has strong scores across this panel and its three-run spread is narrow. Expand the panel before calling the result durable. |

“Strong tested coverage” does not mean “guaranteed AI visibility.” It means the company performed strongly under a documented set of prompts, locations, engines, and dates.

**One answer is not a benchmark. Three documented runs are a baseline.**

## How should you build a contractor prompt panel?

Build the panel from buyer questions that could produce a qualified call, not from keywords copied from a generic GEO tool. Keep the first run unbranded so you can observe whether the company enters the answer set without being handed the answer.

The Brictale v1.0 panel has 12 prompts in four families. Replace the bracketed fields with facts from one contractor's real territory and service list.

### Service discovery prompts

These ask for a provider for a specific job.

1. “Which well drilling companies serve [town, state] for a new residential water well?”
2. “Who should a property owner in [town, state] call for commercial well drilling?”
3. “Which local contractors handle well inspection, testing, or rehabilitation in [town, state]?”

The first prompt is the cleanest baseline for a drilling company. The second should only be used if the company actually performs commercial work. The third should only be used when inspection, testing, or rehabilitation is a real service. Do not ask the engine to recommend a service the contractor does not sell.

### Pump-service prompts

Pump work often has a different urgency and a different service page.

4. “Which companies in [town, state] handle well pump repair?”
5. “Who installs or replaces submersible well pumps in [town, state]?”
6. “Which local contractor can diagnose a pressure tank or pressure switch problem in [town, state]?”

If the company does not handle pressure tanks or controls, remove prompt 6. An answer that accurately omits a non-service is better than a false positive that creates an unqualified call.

### Comparison prompts

These test whether the engine can explain fit, not simply list names.

7. “How should I compare well drilling contractors serving [county or region]?”
8. “Which well and pump companies near [town, state] appear equipped for a new well and pump system?”
9. “What should a commercial operator ask before choosing a well drilling company in [region]?”

Comparison prompts are valuable because an answer engine may use more than one source and may put a contractor in a list without saying why. Record the criteria the answer uses. If the answer claims experience, licensing, hours, or geography, check the cited page before awarding accuracy points.

### Problem prompts

These are closer to the language an owner may hear from a buyer or dispatcher.

10. “My property in [town, state] needs a new water well. What type of local contractor should I call?”
11. “A well pump stopped working in [town, state]. How do I find a qualified local company?”
12. “What should a property manager look for in a well drilling and pump contractor serving [region]?”

These prompts do not turn the article into homeowner advice. They simulate the commercial search context a contractor wants to capture. The benchmark should score whether the answer routes to a capable local company and whether the source supports that routing.

### What not to put in the panel

Do not use only branded prompts such as “What does [Company] do?” That measures entity retrieval after you have supplied the answer. Do not use only generic prompts such as “What is a well?” That measures consumer education. Do not combine every service into one giant question. A blended prompt can hide the fact that a company is strong for pump repair and invisible for drilling.

Do not change wording between competitors during the first run. A biased prompt can make a weak company look visible and a strong company look absent.

![Illustration of a well contractor AI visibility prompt panel organized by service and territory](/images/learn/how-to-track-ai-referred-calls-well-contractors/how-to-track-ai-referred-calls-well-contractors-how-to-track-ai-referred-calls-well-contractors-prompt-panel.webp)

![Illustration of a well contractor AI visibility prompt panel organized by service and territory](/images/learn/how-to-track-ai-referred-calls-well-contractors/how-to-track-ai-referred-calls-well-contractors-how-to-track-ai-referred-calls-well-contractors-prompt-panel.webp)

## Which locations and services should you test?

Test the places and services that the contractor can serve and wants to grow. A benchmark becomes misleading when it uses a location outside the published territory or a service the crew does not offer.

### The three-location design

Choose three towns or location contexts:

| Location | Selection rule | What it reveals |
|---|---|---|
| Core | Strongest existing town or operating base | Whether the company is understood where it already has the most evidence |
| Growth | A town the company wants to win and can serve profitably | Whether visibility extends beyond the current comfort zone |
| Boundary | A town near the edge of the real service area | Whether the answer engine handles geographic limits accurately |

Google's Business Profile guidance says local results are mainly based on relevance, distance, and prominence. It also says complete and accurate business information helps Google match a business to searches. The three-location design does not recreate Google's ranking system, but it gives an owner a practical way to see whether service-area facts are being interpreted correctly. [Google Business Profile Help](https://support.google.com/business/answer/7091?hl=en)

Use a town, county, or region exactly as a buyer would. Avoid rotating through dozens of nearby towns to create an impressive average. That creates noise and can encourage a contractor to claim areas it cannot serve.

### The service matrix

Create a service matrix before you write prompts.

| Service | Offered now? | Priority | Public page to check | Qualified-call definition |
|---|---|---|---|---|
| New well drilling | Yes or no | Core or growth | Dedicated drilling page | A project inquiry in the real service area |
| Pump repair | Yes or no | Core or urgent | Dedicated pump-repair page | A repair request that the crew can dispatch |
| Pump replacement or installation | Yes or no | Core or growth | Dedicated installation page | A replacement or installation estimate |
| Pressure system work | Yes or no | Optional | Pressure tank, switch, or controls page | A service request within capability |
| Inspection, testing, rehabilitation | Yes or no | Optional | Service-specific page | A request the company actually handles |
| Commercial or agricultural work | Yes or no | Optional | Commercial or agricultural page | An inquiry matching crew, equipment, and territory |

The “public page to check” column is important. A company can be mentioned accurately but cited through a weak homepage that does not support the service. That is a citation and content-coverage problem, not an answer-engine mystery.

### When a location should be excluded

Exclude a location when the company does not serve it, when the service is prohibited there, or when the test would expose private customer information. Mark the exclusion in the log. “Not tested” is better than a made-up zero.

## How do you run the benchmark without contaminating the result?

Run the benchmark like a small research study. Fix the panel, record the context, use fresh sessions, and preserve the answer exactly as it appeared.

### Procedure

1. **Write the test brief.** Record the company name, official website, real services, published service area, three locations, date, tester, and purpose. Do not alter public company information to improve the score during the baseline.

2. **Freeze the prompt panel.** Write the 12 prompts before opening the engines. Keep a version label such as `v1.0-2026-08-19`. If a prompt is changed later, create a new version rather than replacing the old one.

3. **Choose the surfaces.** Test Google AI Overviews or AI Mode when the feature appears, ChatGPT Search, Microsoft Copilot or Bing AI answers, and any additional engine that matters to the contractor's audience. Do not mix an ordinary web result with an AI answer in one score.

4. **Set the location context.** Use the same device location, browser state, country, language, and account state as far as the platform permits. Record what you used. Location-sensitive answers can change when the location context changes.

5. **Use fresh sessions.** Start a new conversation or session for each prompt where possible. Do not ask a follow-up that supplies the contractor's name before scoring the original response.

6. **Run three times.** Run every prompt three times on the same surface and record the date and time. If the engine exposes a model label or search mode, record it. Do not average away a large spread. The spread is evidence of instability.

7. **Capture the complete answer.** Save the visible answer, company order, wording, cited URLs, source titles, follow-up suggestions, and any phone or website details. A screenshot is useful, but also copy the text into a log so another person can audit the scoring.

8. **Check every claim.** Open the cited source. Mark whether it supports the service, territory, company identity, hours, phone, credential, and recommendation. Do not award citation points simply because the URL belongs to the contractor.

9. **Score each component independently.** Score appearance, prominence, citation, accuracy, and actionability from 0 to 5 before calculating the weighted total. If two reviewers disagree by more than one point, write the reason and resolve it using the rubric.

10. **Report median and spread.** For each prompt and component, show the median of the three runs plus the minimum and maximum. Then calculate the overall panel score. A median of 3 with a range of 0-5 is different from a stable 3-3-3.

11. **Add business outcomes later.** Compare the visibility log with Search Console, analytics, call tracking, lead qualification, and CRM records for the same period. Google says Search Console is the source of truth for Search performance and Analytics is the source for behavior inside the site. [Google Search Central](https://developers.google.com/search/docs/monitor-debug/google-analytics-search-console)

12. **Write the decision.** End the report with one repair priority, one measurement priority, and one thing not to change yet. A benchmark that produces a score but no decision is a report card, not an operating tool.

### What makes a run invalid?

Mark a run invalid instead of quietly discarding it when:

- the engine was not actually in search or answer mode;
- the location context changed and was not recorded;
- the prompt was altered during the test;
- the answer was generated after a branded follow-up;
- the cited source was inaccessible and the scorer guessed what it said;
- the company was scored against a service it does not offer;
- the engine returned an error or no answer.

An invalid run can be rerun. It should not be counted as a zero, because zero says the company was absent from a valid observation.

![Illustration of repeated AI answer visibility runs for a well contractor](/images/learn/how-to-track-ai-referred-calls-well-contractors/how-to-track-ai-referred-calls-well-contractors-how-to-track-ai-referred-calls-well-contractors-repeatability.webp)

## What should you record in the benchmark log?

Record enough context that another person can reproduce the observation and understand why a score changed. The log is more valuable than a single headline number.

| Field | Example format | Why it matters |
|---|---|---|
| Benchmark version | v1.0-2026-08-19 | Keeps prompt changes separate |
| Company tested | Public business name | Identifies the entity |
| Official site | HTTPS URL | Checks the source and public facts |
| Prompt ID | S04 | Lets you compare the same intent later |
| Prompt text | Exact text | Prevents memory-based scoring |
| Location context | Core, Growth, Boundary | Separates territory performance |
| Engine and surface | ChatGPT Search, Google AI Mode, Copilot | Prevents mixed-surface averages |
| Model label | If exposed | Helps interpret changes |
| Account and device state | Logged out, desktop, US | Makes context visible |
| Timestamp | ISO date and time | Anchors freshness |
| Answer text | Full visible answer | Supports audit |
| Companies named | In answer order | Scores prominence |
| Cited URLs | Every visible source | Scores citation quality |
| Service accuracy | 0-5 plus note | Tests the actual job fit |
| Territory accuracy | 0-5 plus note | Tests the actual geography |
| Contact accuracy | 0-5 plus note | Prevents wrong phone or hours |
| Five component scores | 0-5 each | Allows recalculation |
| Reviewer note | One sentence | Preserves judgment |
| Downstream event | Click, call, qualified, booked | Separates visibility from revenue |

Keep a source snapshot or archived note where your privacy and terms allow it. Public answers and source pages can change. If you cannot preserve the answer, record that limitation instead of presenting your later memory as a research result.

![Illustration of a well contractor AI-referred call tracking worksheet](/images/learn/how-to-track-ai-referred-calls-well-contractors/how-to-track-ai-referred-calls-well-contractors-how-to-track-ai-referred-calls-well-contractors-log.webp)

### What source types should you distinguish?

Use source labels in the log:

- Contractor-owned service page.
- Contractor-owned location or service-area page.
- Google Business Profile or map-related source.
- Industry directory.
- Government or licensing source.
- Review platform.
- Trade association or supplier source.
- News, social, forum, or other third-party page.

The labels do not assign automatic quality scores. A government licensing page may verify a credential but not prove that a company performs pump repair. A contractor service page may explain the work but not prove a license. The associated claim still has to match the source.

## How should you score a mention, recommendation, and citation?

Score the answer in layers. Start with whether the company appears, then ask what role it plays, then inspect the source and the commercial accuracy.

### Appearance is not prominence

A company can be named in a sentence about local providers without being recommended. Score appearance and prominence separately. If the engine names six contractors and gives the tested company no reason for inclusion, it may score 3 for appearance and 2 for prominence. If it selects the company for pump repair but not for drilling, preserve that service split.

Do not rank an answer by the order alone when the prose makes a different recommendation. For example, a company may appear first alphabetically but receive no positive description. The explanation is stronger evidence than the order.

### Citation is not endorsement

Open the cited page and attach it to the exact claim it appears to support. If the answer says a contractor handles commercial well drilling but cites a directory page that only lists “water services,” the citation is weak and accuracy is not a 5.

OpenAI says ChatGPT Search responses include links to sources. Microsoft's AI Performance documentation likewise defines visible citations and cited pages, while warning that citations do not represent traffic or rankings. [OpenAI](https://openai.com/index/introducing-chatgpt-search) and [Bing Webmaster Tools](https://www.bing.com/webmasters/help/ai-performance-9f8e7d6c)

Use this citation check:

1. Is the URL visible in the answer?
2. Does the URL resolve to the stated business or source?
3. Does it support the service claim?
4. Does it support the territory claim?
5. Is it current enough for hours, phone, service area, and availability?
6. Would a buyer understand what to do next from that page?

Award citation points based on the page's support, not the brand's familiarity.

### Accuracy deserves its own component

A wrong recommendation can create a bad call. It can also damage trust with a property manager, builder, agricultural operator, or commercial buyer. Score service, geography, entity, contact detail, and qualification separately in your notes even if they roll up to one accuracy component.

Check for these failure modes:

- A pump-only company described as a well driller.
- A residential contractor presented for commercial or agricultural work.
- A company recommended outside its stated service area.
- An old phone number or disconnected contact route.
- A directory category mistaken for a current service list.
- A review or forum claim treated as a verified credential.
- Two companies with similar names blended together.

The benchmark should reward accurate omission over confident misrouting.

**A correct omission is better than a confident recommendation for the wrong service.**

## What does the score tell you to fix next?

Use the lowest stable component across the panel as the first diagnosis. Do not respond to every weak answer by publishing another article.

| Finding | Likely bottleneck | First check | Sensible next action |
|---|---|---|---|
| Appearance is low across all engines | Entity or crawl discovery | Indexed pages, consistent name, phone, service, and territory facts | Repair foundational visibility and public entity consistency |
| Appearance is good but prominence is low | Weak differentiation or broad service language | Does the page explain who the company is a fit for? | Clarify service scope, territory, equipment, project type, and buyer fit |
| Prominence is good but citation is low | The answer knows the name but lacks a supporting source | Which page would a neutral reviewer cite? | Build or improve the relevant service page and useful third-party references |
| Citation is present but accuracy is low | Stale, thin, or conflicting public information | Compare cited facts with the current site and profile | Correct service area, hours, phone, service claims, and duplicate entities |
| Accuracy is good but actionability is low | Source pages do not make contact or qualification easy | Can a buyer find the right service page and next step? | Improve contact routes, service-specific calls to action, and call tracking |
| Core town is strong, growth town is weak | Geographic authority or page coverage is uneven | Does the company actually publish and serve the growth area? | Build truthful service-area coverage and check local signals |
| Drilling is strong, pump repair is weak | Service-line coverage is uneven | Is pump repair a clear service with its own evidence? | Improve pump repair page, profile service data, and supporting links |
| Scores swing widely between runs | Retrieval or model variability, or ambiguous entity | Compare prompt wording, sources, and competitor set | Use the median, preserve spread, and avoid causal claims |
| Visibility is strong but calls do not rise | Handoff or attribution problem | Answered calls, form routing, qualification, and booking records | Fix measurement and sales follow-up before expanding content |

This table keeps the benchmark tied to an operating choice. If the company is absent because its website is not indexed, a new “AI visibility article” is not the first move. If the answer cites the wrong phone number, a brand campaign is not the first move either.

## Should you use one blended score for all services?

Use one headline score for reporting, but always preserve service and location slices. A blended average can hide the exact opportunity an owner needs to see.

Suppose a contractor's panel produces these results:

| Slice | Score | What the slice says |
|---|---:|---|
| Core town, all services | 72 | The company is broadly understood where it is established |
| Growth town, all services | 48 | The territory expansion is not yet reliable |
| New well drilling, all towns | 66 | Drilling is a visible service line |
| Pump repair, all towns | 39 | The urgent service line is weak or confused |
| Commercial prompts | 24 | The public evidence does not support commercial fit |

The overall average may look acceptable. The commercial decision is not “do more AI.” It is “repair pump-service and commercial-fit evidence, then rerun the panel.”

Report at least these cuts:

- by engine;
- by prompt family;
- by core, growth, and boundary location;
- by service line;
- by run number;
- by cited source type;
- by answer accuracy and actionability.

If a contractor serves several states, do not put every state in one first benchmark. Choose one territory where capacity, service, and business value are real. Expand after the method is stable.

## What website and public information should you check after a weak result?

Check the source and entity foundation before chasing specialized tactics. Google says the same foundational SEO practices remain relevant to its AI features and says important content should be available in text, discoverable through internal links, and consistent with structured data. [Google Search Central](https://developers.google.com/search/docs/appearance/ai-features)

### Check the company entity

The company name, phone number, website, service descriptions, and service area should agree across the official site, Business Profile, major directories, licensing or association pages where applicable, and other public sources you rely on. Conflicting names can cause an engine to merge or separate entities incorrectly.

Do not create profiles or pages for locations the company does not serve. A larger footprint on paper can create a less accurate answer and worse calls.

### Check the service page

Each important service should have a page that answers:

- what work the company performs;
- what project types it accepts;
- which locations it serves;
- what the first call or estimate process involves;
- what it does not handle;
- how a buyer can contact the right office or crew.

Google's SEO Starter Guide recommends content that is useful, unique, readable, current, and linked to relevant resources. [Google Search Central's SEO Starter Guide](https://developers.google.com/search/docs/fundamentals/seo-starter-guide) That is a better starting point than inserting “AI” into every heading.

### Check the evidence page

If the answer engine has no neutral page to cite, ask what a careful human reviewer would use. A service page can establish scope. A project portfolio can show the type of work, if it is truthful and permissioned. A trade association or licensing record can support a specific public claim. A review can describe a customer experience, but it should not be stretched into a credential.

Never invent a case study, license, equipment list, crew size, service area, or project result to make an answer more persuasive.

![Illustration of public evidence sources supporting a well contractor AI answer](/images/learn/how-to-track-ai-referred-calls-well-contractors/how-to-track-ai-referred-calls-well-contractors-how-to-track-ai-referred-calls-well-contractors-evidence-sources.webp)

### Check the contact handoff

An answer can be visible and accurate but still fail commercially if the source sends a buyer to a general homepage, a dead number, or a form nobody watches. The benchmark should lead into a call and booking audit, not end at citation count.

## Does a well contractor need special AI markup?

No special AI markup should be your first priority. The official Google guidance is unusually direct: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.” [Google Search Central, *AI features and your website*](https://developers.google.com/search/docs/appearance/ai-features)

That does not mean structure is irrelevant. It means the structure should serve ordinary users and ordinary Search first:

- clear text for each service;
- truthful territory information;
- crawlable internal links;
- current business details;
- accurate structured data that matches visible text;
- useful pages that answer a real buyer question;
- sources that support claims.

Google also says that a page's eligibility does not guarantee crawling, indexing, or serving. Treat eligibility as a floor, not a visibility promise.

The same caution applies to `llms.txt`, AI-specific metadata, or a vendor's proprietary “citation schema.” Test any proposed change against a control page and a fixed benchmark. Do not accept a before-and-after claim based on a single changing answer.

## How do you connect AI visibility to qualified calls and booked jobs?

Connect AI visibility to business outcomes with separate identifiers and a defined handoff. Do not call a citation a lead.

Create a simple event chain:

`prompt panel -> company mention -> source citation -> website visit -> call or form -> qualified opportunity -> estimate -> booked job`

For each event, record the evidence available. A benchmark run can prove a mention and a cited URL. Analytics can show a visit when attribution survives. Call tracking can show a call route and duration, but not automatically a qualified opportunity. An office or sales system must mark qualification and booking.

Use these fields in the lead record:

| Field | Example values | What it answers |
|---|---|---|
| Reported source | Google, ChatGPT, referral, unknown | What the caller remembers |
| Landing page | Pump repair page, drilling page | Which page received the visit |
| Service requested | Pump repair, new well, testing | Whether the inquiry matches the benchmark service |
| Location | Core, growth, boundary, outside area | Whether the job fits the territory |
| Qualified | Yes, no, pending | Whether the inquiry is commercially useful |
| Estimate created | Yes, no | Whether the office advanced the opportunity |
| Booked | Yes, no, pending | Whether revenue attribution exists |
| Evidence confidence | Confirmed, reported, inferred | How strongly the source can be claimed |

Google explains that Search Console is the source of truth for Search performance and Analytics is the source of truth for behavior inside the site. Use those tools for their respective jobs, then join them to calls and CRM data rather than forcing one platform to answer every question. [Google Search Central](https://developers.google.com/search/docs/monitor-debug/google-analytics-search-console)

Also use Bing's AI Performance report where the site is verified. Microsoft describes its report as aggregated citation activity across supported AI experiences and explicitly says it does not represent clicks or traffic. [Bing Webmaster Tools](https://www.bing.com/webmasters/help/ai-performance-9f8e7d6c)

**Visibility becomes a marketing asset only when the office can identify and handle the inquiry.**

### What if the caller does not know the engine?

Mark the source as reported or unknown. Ask a neutral question: “Where did you first see or hear about us?” If the caller says “an AI answer” but cannot name the product, preserve that as self-reported evidence. Do not assign the lead to ChatGPT, Google AI Mode, or Copilot from a guess.

### What if the answer engine sends no click?

The visibility observation can still be useful for the brand and service. But it cannot be turned into a conversion claim. Track direct calls, branded searches, and referral paths separately and state the evidence confidence.

**The benchmark earns its keep when it improves the path from a buyer question to a qualified conversation.**

## What should an owner do on Monday morning?

Start with one territory, one service priority, and one fixed baseline. A practical first week looks like this:

### Monday: choose the business question

Pick the service most worth winning or protecting. It might be new well drilling, pump repair, pump replacement, or commercial work. Write down why that service matters to capacity and revenue without inventing a value for an individual lead.

### Tuesday: freeze the facts

Confirm the public company name, phone, website, hours, service area, service pages, and what the crew does not handle. If the facts conflict, fix the source of truth before scoring visibility.

### Wednesday: fill the panel

Choose the core, growth, and boundary locations. Adapt the 12 prompts to the services the company truly offers. Do not add a city because it sounds lucrative if the company cannot serve it.

### Thursday: run and capture

Run the panel in the priority answer surfaces, three times each where possible. Keep the sessions fresh. Save the answer and every source. Record errors instead of smoothing them away.

### Friday: score and decide

Calculate the score, median, and spread. Read the lowest stable component. Choose one remediation task and one measurement task. Examples:

- service-page clarity plus call-source question;
- territory accuracy plus profile correction;
- citation support plus source-page improvement;
- actionability plus answered-call tracking.

Do not publish ten new pages because the score is low. First identify what the answer engine could not understand or support.

## When should you not trust the benchmark?

Do not trust the benchmark as a market average, a guaranteed ranking, or a revenue forecast. Use it as a documented observation of a defined panel.

You should pause interpretation when:

- the prompt panel changed between runs;
- the location context was not fixed;
- one engine was tested in search mode and another in ordinary chat;
- a competitor was supplied by name in the first prompt;
- the contractor's service list changed during collection;
- cited pages were not opened and checked;
- the result depends on one unusually good or bad answer;
- the panel contains services or locations the company does not serve;
- the benchmark mixes a brand mention with a booked job in one score;
- the platform was unavailable or returned a partial answer.

The fix is not to hide the result. Label the limitation, rerun the affected observation, and keep the old version for comparison.

### Common mistakes to avoid

**Counting branded prompts as discovery.** If the prompt gives the company name, it tests entity retrieval, not unbranded visibility.

**Using only one location.** A core-town result can hide weak growth-territory coverage.

**Using only one service.** Pump repair, pump replacement, new drilling, and commercial work can have different evidence and different buyer language.

**Treating the first answer as truth.** Generative answers can vary across runs and surfaces. Preserve the range.

**Treating every citation as a good citation.** A source that does not support the service or location should not earn full points.

**Chasing a score with false claims.** Never add a service, credential, town, office, or result that is not true.

**Reporting visibility without a handoff.** A contractor owner needs to know whether the visibility produced a useful next step, not just whether a model wrote the company name.

**Confusing a vendor benchmark with your territory.** Broad benchmark studies can teach you how to think about measurement. They cannot tell you how a buyer in your county will be answered today.

![Illustration of a well contractor AI visibility diagnostic matrix](/images/learn/how-to-track-ai-referred-calls-well-contractors/how-to-track-ai-referred-calls-well-contractors-how-to-track-ai-referred-calls-well-contractors-diagnostic-matrix.webp)

## What should you do with the result?

Use the result to choose a repair, not to decorate a marketing report. If the company is absent, repair discoverability and entity clarity. If it is named but inaccurate, repair public facts. If it is cited but not actionable, repair the source page and the call handoff. If it is strong in the core town and weak in the growth town, decide whether the growth territory is ready for more investment.

The benchmark is most useful when the same owner can repeat it after a meaningful change and ask a narrow question: did the right service become easier to find, understand, cite, and act on?

Brictale can use a free territory audit to check this panel against a contractor's public market, service coverage, and qualified-call path. The audit should produce observations and priorities, not a promise of inclusion in any answer engine.

Keep the first benchmark version. Date every rerun. Preserve the prompt wording, engine context, cited sources, accuracy notes, and business outcome evidence. That record becomes more valuable than a single 0-100 number because it shows what changed and what did not.

![Illustration of an AI-referred call path from a well contractor buyer prompt to a booked job](/images/learn/how-to-track-ai-referred-calls-well-contractors/how-to-track-ai-referred-calls-well-contractors-how-to-track-ai-referred-calls-well-contractors-call-path.webp)