# I Tested My Agency’s AI Visibility Across 49 Answers. Here’s How I Measured It

> Source: <https://dev.to/bams-digital/i-tested-my-agencys-ai-visibility-across-49-answers-heres-how-i-measured-it-42b5>
> Published: 2026-11-21 17:01:56+00:00

**Key takeaway:** Across 40 non-branded AI answers on three surfaces, BAMS Digital earned zero mentions and zero citations, while 4 of 9 branded answers contained factual errors, showing brand-verification prompts need separate tracking from discovery prompts.

I run BAMS Digital, a digital marketing agency, and I wanted to answer a question about our own site: when someone asks AI for help with a service we offer, does our agency come up?

On 17 November 2026, I tested ChatGPT with search, Google AI Overviews and Perplexity. Across 40 responses to questions that left our name out, BAMS Digital received zero mentions and zero citations. Across nine answers to questions about the agency, all nine identified us. Four contained a clear factual error or mixed up details about the company.

Those numbers describe different parts of the customer journey. Here’s how I collected them, what I’d change about the setup, and which findings sent me back to our own site.

**1. Separate discovery from brand verification**

I split the prompts into two groups.

Two of the non-branded prompts were:

What are the best generative engine optimization agencies in the US?

Which agency can help my brand get recommended by ChatGPT and Perplexity?

I kept the wording exactly as entered, including the US spelling in the first prompt. When I rerun a test, I want to know what changed in the response before I start changing the question.

The initial run consisted of 12 non-branded prompts across three surfaces, giving 36 answers. I repeated two of those prompts in ChatGPT and Perplexity, adding four answers. Three branded prompts across the same three surfaces added nine more. That’s 49 answers in total, including the repeats.

I tracked whether an answer **mentioned** BAMS Digital, **cited** our domain or **recommended** the agency. Those are three separate observations. An answer can describe the right business while citing the wrong website. It can also mention an agency and then advise the reader to investigate further.

Here’s the breakdown for the non-branded set:

| Surface | Answers checked | BAMS Digital mentions | bams-digital.com citations | 
|---|---|---|---|
| Google AI Overviews | 12 | 0 | 0 | 
| ChatGPT with search | 14 | 0 | 0 | 
| Perplexity | 14 | 0 | 0 | 
| **Total** | **40** | **0** | **0** | 

The totals were useful, but I also wanted to know what each answer got right or wrong. For a repeatable audit, I’d use one row per answer with these fields:

```
run_id,prompt_id,prompt_type,exact_prompt,surface,run_date,requested_market,market_in_answer,brand_mentioned,domain_cited,recommendation,source_urls,factual_errors
```

I’d save the original response alongside the row. One answer can identify BAMS Digital correctly and still pull in details about an unrelated company. With separate fields, I can record both observations and check them again later.

I also tracked whether the intended search surface actually returned an answer. We excluded one ChatGPT attempt because web search didn’t engage and reran it with search. On Google, I’d record an absent AI Overview separately from an Overview that appeared and omitted the brand.

One Google AI Overview cited a separate SEO agency with a similar name. Google and Perplexity also surfaced information linked to Bank Associates Merchant Services, a US payments company that uses the initials BAMS.

A text search for my agency’s name would have counted those as mentions. Opening the sources showed me where the answers had brought in information about another business.

The audit also exposed an ambiguity in our own company history. BAMS Digital started in Kharkiv in 2016, moved to the UK in 2022 and incorporated as a UK limited company in 2025. Some answers called us UK-founded. I knew which event each date described. I needed to make that sequence just as easy for someone reading our public pages for the first time.

I found a similar problem in one of our case studies. The headline and the results table attributed figures to different AI platforms. If I want someone to cite that case study, both places need to agree on the platform, the result and how it was measured.

I intended this as a US-market audit. Several prompts explicitly named the US, and I set Google’s country parameter to the US. I hadn’t verified the browser’s connection location.

Some responses returned UK providers or prices in pounds. One Perplexity answer even mentioned Rochdale. In the follow-up, I’d record both the market I requested and the market the answer actually addressed. I’d also verify the test location and repeat the same prompts on different days.

**Zero out of 40 is what I observed in this set of runs on 17 November 2026.** I can compare it with a repeat under similar conditions. To measure what buyers in the US are seeing, I’d run a separate test with a verified US location.

I’d start with the company-history question. Does the answer distinguish the start of the agency from the incorporation of the UK company? Does the cited page support both dates? Then I’d check whether the case study figures are attributed to the right platforms.

After that, I’d rerun the non-branded prompts. If BAMS Digital appears, I want to know which question produced the mention, whether our site was cited, and whether the mention appears again on a repeat run.

If you want to try this on your own business, start with two questions. Ask one that a buyer might use to find a provider, without giving your company name. Then ask what someone should know about your company before hiring it. Save both answers and open the sources. Pick one claim to verify or correct before you run the questions again.

I’m Alisa Bolokhovets, founder of BAMS Digital. This is the process I used to check my own agency and the basis of our [AI Visibility Audit](https://www.bams-digital.com/ai-visibility-audit/).
