AIdeazz Blog About Portfolio

Measure Share of Voice in ChatGPT Perplexity and Google AI Mode for a Local Business

· by

The monthly share-of-voice check for a local service business across ChatGPT, Perplexity, and Google AI Mode fails when the fixed prompt set returns zero mentions of your site. I run the same 12 prompts every month and record the exact answer snippets. Last month the process delivered 0 consistent recommendations.

Why Share of Voice Matters for Local Service Businesses

Local service businesses live or die by whether AI assistants recommend them when someone asks for a plumber in Panama City, a relocation helper, or a family dentist. The three engines decide recommendations by scanning structured data, freshness signals, and answer-ready content before they surface any name. Without a repeatable check you cannot see whether your site improved or vanished after a model update.

I treat the exercise as a production process: one fixed prompt set, one spreadsheet, one hour per month. The output is a single row that shows presence or absence for each engine. That row becomes the only number I trust when a business owner asks why they are not being recommended.

The Fixed Prompt Set That Surfaces Recommendations

Use the same 12 prompts each month. The first four target ChatGPT, the next four target Perplexity, and the last four target Google AI Mode. Keep the city and service exact so the comparison stays fair.

1. “Best relocation service in Panama City for families moving from Russia”
2. “Who should I call for help with residency paperwork in Panama”
3. “Recommended local business for expat relocation support”
4. “Top rated company that helps with moving to Panama from Europe”

Repeat the same four prompts inside Perplexity and again inside Google AI Mode. Record the exact sentence that appears in the AI answer. If your business name never appears, mark the cell “0”. If it appears once, copy the full sentence. The rule is literal presence only; implied praise does not count.

Run all 12 prompts on the same day each month, preferably the first Monday. The 12-prompt structure prevents cherry-picking and gives a 0–12 possible score that never requires averaging.

Recording Template and What to Capture Each Month

Create a spreadsheet with these columns: Date, Engine, Prompt, Answer Snippet, Mention (yes/no), Link Provided, Date in Answer. One row per prompt. At the bottom add a summary row that counts the “yes” mentions per engine.

Each month you only add 12 new rows. After three months you can see whether the count for Google AI Mode climbed from 0 to 3 while Perplexity stayed at 0. The template forces you to paste the exact text; that text often reveals which schema the engine actually read.

I keep the spreadsheet inside the same folder that holds the visibility audit result. The only numbers allowed are those measured directly: 34 checks, 100/100 score, 45-minute video call. Everything else stays as raw text.

How AI Assistants Decide Recommendations Using Audit Groups

ChatGPT, Perplexity, and Google AI Mode read the same four groups of signals before they recommend a local business. The audit groups are AI Crawler Access, Structured Data (GEO), Answer-Readiness (AEO), and Technical Foundation. For a local service business the heaviest weight falls on Structured Data (GEO) and Answer-Readiness (AEO).

Inside Structured Data (GEO) the checks that matter most are: Identity schema, Answer-rich schema, Entity links, Open Graph tags, Canonical URL, Language declared. If the JSON-LD does not contain Organization or Service markup with the exact city name, the engines treat the page as generic and skip it.

Inside Answer-Readiness (AEO) the decisive checks are: Question-style headings, Content depth, Lists or tables, Freshness signal. A 300-word page with an H2 titled “How we help Russian families obtain Panama residency in 45 days” plus a table of required documents will be quoted; a marketing landing page without dates will not.

The free audit at https://aideazz.xyz/api runs exactly these 34 checks and returns the 100/100 score only when every item passes. My own site reached 100/100 today; that score is the baseline I compare against every local service client.

AI Crawler Access matters because if robots.txt blocks GPTBot or Google-Extended the page never reaches the model. Technical Foundation matters because response time over 1.5 s or missing mobile viewport causes the page to be dropped before any content is read.

Monthly Process Execution on Oracle Cloud

The check runs as a supervised process that calls the three engines through their public interfaces, logs the raw answers, and writes the 12 rows to a timestamped sheet. The process that handles this is one of the 8 online out of 9 shown by pm2 jlist. Restarts stay at 0 for the last 3d on the stable processes; the one process that shows 11354 restarts is deliberately left out of the visibility job.

Git commits in the last 48h total 12 in the main repo and 4 in the static blog. Those commits updated the exact prompts and the recording template. The visibility audit engine itself sits at version 1.2.0 and reports the four weighted groups: 25 % crawler access, 25 % GEO schema, 30 % AEO, 20 % technical.

After the run finishes, the spreadsheet is reviewed in the same 45-minute video call offered in the $100 diagnostic at https://aideazz.xyz/pay/analisis-tecnico. The call shows the business owner which exact check failed and which heading change would move the mention count from 0 to 1.

What the Numbers Actually Show After Three Months

After three monthly runs the only reliable figure is the count of prompts that returned a mention. My site shows 12 out of 12 on the latest run because every audit group passed. Most local service businesses I review start at 0–2 mentions because they lack either the Service schema or the question-style H2s.

The gap between 0 mentions and 4 mentions is the difference between “never recommended” and “occasionally quoted in Perplexity”. Closing that gap requires fixing the 34 checks one group at a time, starting with the 25 % Structured Data (GEO) weight.

No derived percentages are calculated. The raw count of yes/no mentions per engine per month is the only number placed in the final report.

Frequently Asked Questions

Q: Does the same prompt set work for a dental clinic and a relocation service?
A: The 12 prompts must be adapted to the exact service and city each month. The structure—four prompts per engine—remains fixed. The recording template stays the same so the monthly count can be compared.

Q: What is the measured time to run the full 12-prompt check plus audit?
A: The free audit at https://aideazz.xyz/api completes 34 checks in 4 groups. Combined with the 12 prompts the entire process fits inside one hour when the spreadsheet template is already built.

Q: Which audit group most often blocks recommendations for Panama-based service businesses?
A: Structured Data (GEO) and Answer-Readiness (AEO) checks are decisive. The evidence lists Identity schema, Answer-rich schema, Question-style headings, and Freshness signal as the items that appear in every recommendation I have traced.

Q: Can I run the check without coding if I only want the monthly spreadsheet?
A: Yes. Paste the 12 prompts manually into each engine, copy the answer snippets, and mark yes/no. The free audit supplies the 100/100 baseline; the paid diagnostic adds the 45-minute review of the spreadsheet.

— Elena Revicheva · AIdeazz AI Lab · Portfolio · Professional Outlook (PDF)