AI assistants recommend firms that other sites already discuss. Most B2B firms give them little to work with: few third-party mentions, a name the engine confuses with a namesake, and no pages that match the buyer's question. When we tested our own firm, Perplexity named us in 0 of 42 answers to unbranded buyer questions (our data, October 2026).
What happened when we tested our own firm?
We came out close to invisible. We wrote 16 questions that a buyer of outbound or go-to-market help would ask, then ran each one three times through Perplexity's Sonar API, for 48 answers (our data, October 2026). Fourteen were unbranded, such as "What is GTM engineering and which firms offer it as a service?" and "Can AI agents replace SDRs for B2B outbound?"
The results:
- Exponential Advisory appeared in 3 of 48 answers. All three came from the one question that contained our name.
- We appeared in 0 of the 42 answers to unbranded questions.
- Our domain was cited as a source in 0 of 48 answers.
A second question asked who helped FastLane grow its market share from 10% to 75%. That is our own result, and Perplexity did not name us in any of the three runs. It cited FastLane's site instead.
This covers one engine on one day. Other assistants will answer differently. We measure every change against these numbers, and we have no gain to report yet.
Why does AI describe a different company when you ask for yours by name?
Usually another entity with a similar name has a clearer footprint on the web. When we asked "What is Exponential Advisory?", Perplexity described a different firm with a similar name, an HR and training consultancy, and cited that firm's site (our data, October 2026). Entity confusion is the first problem we are fixing. Until it is fixed, any visibility we gain could send buyers to the other firm.
Fix it in this order:
- Use one exact firm name, one category line and one description on your site, your LinkedIn company page, Crunchbase and every directory profile you keep.
- Put your domain next to your name in every profile and byline you control.
- Say what you sell, and to whom, in the first lines of your home page and About page.
- Add Organization markup with sameAs links to those profiles. It helps Google connect them. Expect little from it in AI answers: an Ahrefs controlled test of 1,885 pages that added schema found no meaningful lift in AI citations (Ahrefs, May 2026).
- Ask the branded question again each week until the answer describes your firm.
How do you measure AI visibility without fooling yourself?
Use a fixed set of prompts, run each one several times, and report a rate. One answer tells you almost nothing. SparkToro and Gumshoe collected 2,961 AI responses from 600 volunteers and found below a 1-in-100 chance that the same list of brands appears twice (SparkToro, January 2026).
Set up the panel like this:
- Write 15 to 45 buyer questions across five buckets: category, comparison, problem, cost and branded.
- Freeze the wording for at least four weeks.
- Run each question at least three times in fresh sessions with web search on.
- Log for every answer whether the firm was named, whether the domain was cited, which domains were cited, and whether the description was correct.
- Report each number as a rate across all runs, with branded and unbranded questions kept apart.
| Metric | What it counts | Our baseline (October 2026) |
|---|---|---|
| Unbranded mention rate | Answers that name the firm, out of all unbranded answers | 0 of 42 |
| Branded mention rate | Answers that name the firm when the question names it | 3 of 3 |
| Citation rate | Answers that cite the firm's domain as a source | 0 of 48 |
| Description accuracy | Branded answers that describe the right firm | Wrong firm described |
| Source map | Domains cited across all answers, with counts | LinkedIn first, 36 citations |
Keep mentions and citations in separate columns. An engine can cite a page without naming the firm behind it, so the two numbers move separately.
Where does AI look when it recommends a firm?
Mostly on other people's sites. Foundation and AirOps analysed 57.2 million citations across 50 B2B brands from December 2025 to February 2026. On unbranded questions, 2.24% of citations went to the brands' own sites (Foundation x AirOps, 2026). Across 75,000 brands, Ahrefs found that web mentions correlated 0.664 with visibility in Google's AI Overviews, against 0.218 for backlinks (Ahrefs, 2025). Both studies are correlations, and both measures grow with brand size. They still point the same way: third-party pages carry most of the weight.
Our own source map shows where that happens for go-to-market questions (our data, October 2026). LinkedIn was the most cited domain, with 36 citations across 7 of our 16 questions. Next came the blogs of outbound agencies and data tools, crypto media such as CryptoSlate, The Block and PANews, and consulting firms such as Deloitte and McKinsey. Read the list as a map of where the engine looks for this topic.
Turn the map into a target list:
- Post original findings on LinkedIn under a named person, each with one number from your own data.
- Pitch the same findings, with the method and date, to the trade and crypto publications in your map.
- Ask partners and clients for a listing or a co-published case study.
- Answer journalist requests in your niche with two or three quotable sentences and one number.
What pages should you build on your own site?
Build pages whose title is the buyer's question. Your site still feeds live retrieval, and engines cite the retrieved pages that match the prompt. Ahrefs studied 1.4 million ChatGPT prompts and found a title-to-prompt similarity of 0.602 for cited pages, against 0.484 for pages that were retrieved and then left out (Ahrefs, April 2026). Pages with natural-language slugs were cited 89.78% of the time, against 81.11% for the rest.
For each page:
- Use the literal question as the title and the slug.
- Answer it in the first paragraph, in under 60 words.
- Use the buyer's next questions as headings.
- Include at least one fact the reader cannot find elsewhere, with its method and date.
Pick narrow decisions your buyers make, such as how fast verified email lists decay or what to pay per qualified meeting. Broad category pages compete with every large publisher.
Do llms.txt files and schema markup help?
Neither has shown a measurable effect on citations. SE Ranking studied about 300,000 domains and found no relationship between having an llms.txt file and being cited by AI (Search Engine Journal, November 2025). The Ahrefs schema test found no meaningful lift either. Add an llms.txt file if you like, since it costs nothing, and keep schema for Google's rich results. Do not pay anyone to optimise either one.
One technical check does matter. Your robots.txt file and your CDN must let AI crawlers in. OpenAI says a site must allow its OAI-SearchBot crawler to appear in ChatGPT search results (OpenAI). Check your server logs for that crawler and for PerplexityBot.
To get your growth plan and agent stack mapped on your own data in two weeks, start with The Diagnostic at $2,500.