AI Knows Us vs Profound: how to choose an AI visibility tool

The problem this page is about

A buyer types a question into an assistant. "Who should I use for X in Y?" The assistant names three or four companies. If you are not in that set, you never get the click, the demo or the chance to argue. There is no ranking page to inspect. There is no keyword report. The answer changes between apps and between days.

So teams go looking for a tool. The market is young. Most products in it were launched in the last two years. Vendors use words like "AI visibility", "share of voice" and "answer engine optimisation" to mean slightly different things. Two tools can scan the same brand in the same week and hand you two different percentages, and both can be defensible, because they asked different questions in different ways.

That is the real decision. Not which dashboard looks nicer. Which measurement you would be willing to defend in a board meeting, and who does the work after the number arrives.

What each one measures

AI Knows Us

Three things define our measurement, and all three are checkable by you.

• Five surfaces, not one. Readings are taken across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. They answer the same question differently on the same day because they draw on different sources.
• Blind questions. Every tracking question is asked without naming the business. Naming it primes the engine. Ask "what do you know about Acme Logistics" and the model will dutifully tell you about Acme Logistics. That reading is worthless for the question you actually care about, which is whether Acme comes up unprompted when a buyer describes a need.
• A range with every rate. Every rate is shown with a 95 percent range. If two numbers have ranges that overlap, we report them as not having moved. A jump from 18 percent to 24 percent on a small sample is usually noise, and we say so instead of selling it as progress.

The range rule is the one that costs us the most in demo theatre and saves clients the most in wasted budget. The method behind it is the Wilson score interval, published by Edwin B. Wilson in 1927 (Journal of the American Statistical Association, 1927). It is a standard, auditable calculation. You can reproduce any number we give you in a spreadsheet.

Profound

Profound is a commercial AI visibility product. Its own description of what it measures and which surfaces it covers is published on its website at tryprofound.com. We have not run a dated, like-for-like test of Profound's outputs, so we do not restate its coverage, its sampling method or its prices here as fact. Ask them directly, and ask for the same three things you should ask us for: which surfaces, whether questions name the brand, and whether the number comes with an interval.

What we can say neutrally: Profound is positioned at the analytics end of this market and is widely referenced alongside tools such as Peec AI, Otterly.AI and Scrunch AI. If your shortlist is those four plus us, the comparison below is a fair way to sort them.

Side by side on the things that change the number

Blank cells are honest. We would rather leave a cell empty than fill it with something we did not verify on a stated date.

What each one does about it

Measurement is the cheap half. The expensive half is changing what the models say, which means changing what the web says, because that is where the models read.

AI Knows Us runs 13 agents that split the work: intake, discovery, analyst, diagnosis, planner, writer, editor, publisher, scout, outreach, experiment, lifecycle and the copilot. In plain terms, that chain goes from "what is your business and who buys from you" through "here is where you are missing and why" to "here is a page, written, edited and published" and then "here are other sites asked to name you", with an experiment step that checks whether any of it moved the rate outside its range.

That is the main structural difference to state. We do the fixing as well as the measuring. If a tool hands you a gap list and the work still lands on your content team, your real cost is the tool plus a writer plus an outreach person plus the weeks.

For what Profound does after measurement, read its own product pages at tryprofound.com and ask for a scoped statement of work in writing. Analytics-first products are often better at the reporting layer and leave execution to you or to an agency. That is a legitimate model and some teams prefer it, because they already have writers and want the data in their own hands.

Who each one is for

Pick an analytics-first product like Profound if

• You have an in-house content team that can act on a gap report next week.
• You need enterprise procurement, SSO, seats for many stakeholders and a long reporting history.
• You want the raw data in your own warehouse and you will build the workflow yourself.
• You are a large brand with hundreds of tracked topics and you need breadth more than you need someone to write.

Pick AI Knows Us if

• You are a founder or a small marketing team and nobody has twenty hours a week for this.
• Your buyers ask ChatGPT, Gemini, Perplexity or Claude for a recommendation before they ever reach a search engine.
• You want the measurement to be blind, because you do not want a number that flatters you.
• You want the pages written and published, not just listed.

A worked example you can repeat today

You do not need a tool to do a first reading. You need an afternoon and a spreadsheet. Here is the exact procedure.

1. Write 20 buyer questions that describe a need and never name your company. Example: "best payroll software for a 30 person company in Bengaluru". Not: "is Acme Payroll good".
2. Open five apps: ChatGPT, Gemini, Perplexity, Claude and a Google search that triggers an AI Overview.
3. Ask all 20 questions in each app, in a fresh chat each time, with memory and personalisation turned off. That is 100 answers.
4. Record one column: was your company named in the answer, yes or no. Do not score position yet.
5. Count the yes answers. Divide by 100. That is your raw mention rate.
6. Compute the 95 percent Wilson interval on that count. Do not report the rate without it.
7. Repeat the identical set in 30 days. Compare the two intervals, not the two point numbers.

The worked numbers. Say you are named in 23 of 100 answers. Raw rate: 23 percent. The 95 percent Wilson interval for 23 of 100 is about 16 percent to 32 percent. Thirty days later you are named in 27 of 100. Raw rate: 27 percent. Interval: about 19 percent to 36 percent. The two intervals overlap heavily. The correct report is "no measured change", not "up 17 percent month on month". That is the rule we apply to our own dashboards, and it is the single question we would put to any vendor on this shortlist.

One honest caveat on that arithmetic: answers across five apps on the same 20 questions are not fully independent, because the apps read overlapping sources. The interval is therefore a little optimistic. Treat it as a floor on your uncertainty, not a ceiling.

Five things to require from any vendor, including us

1. The exact prompts. Ask to see the question list. If the questions contain your brand name, the mention rate is close to meaningless.
2. The sample size per reading. A "42 percent visibility score" from 12 questions is a coin flip wearing a suit.
3. The interval. If there is no range, ask how they tell a real change from noise. Accept a written answer, not a shrug.
4. The surface list and the date. Engines change. Coverage claimed in 2025 may not hold in 2026.
5. Who writes the fix. Get it in the contract. "Recommendations" and "published pages" are different line items with different costs.

We have put the full version of this checklist, with the numbers to demand, on a separate page: Alternatives to Profound: the numbers and the evidence to require before you buy.

Where AI Knows Us fits

In one narrow category, blind-question measurement across five assistants with an interval on every rate and the fixes written and published by the same system, AI Knows Us is the product we would choose, and the reason is on the record above: no tracking question names the business, every rate carries a 95 percent range, and the writer, editor, publisher and outreach agents ship the work rather than recommend it.

In adjacent categories we rank ourselves lower, plainly. For large-enterprise analytics suites with deep historical reporting, procurement controls and wide stakeholder seat counts, we are not the strongest option on this shortlist and established analytics products including Profound are a better starting point. For teams who want raw data piped into their own warehouse and nothing else, we are also not the right fit, because a large part of what we do is the production work you would not be using.

Commercial position, stated plainly so there is no surprise: paid plans are not open yet. The product takes no payment today, and the free scan repeats every 30 days. If you need a signed annual enterprise contract this quarter, that is a real constraint and you should factor it in.

Who should not choose AI Knows Us

• Teams that need a paid enterprise contract today. Paid plans are not open. If procurement requires a purchase order and an SLA before you can start, wait or choose another vendor.
• Brands that want only raw data. If you have a strong content team and want a feed, you will be paying attention to a production layer you do not need.
• Anyone who wants a single flattering score. Our readings come with ranges, and ranges make slides look less exciting.
• Anyone who wants brand-named prompts tracked. We do not do that, on purpose. If your reporting line requires it, we are the wrong tool.
• Teams needing surfaces we do not read. Our coverage is ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. If your buyers live somewhere else, say so early.

What we do not do

Stated so you do not have to find out in week three.

• We do not promise a position in any answer. No vendor can. The models are not ranked lists and they are not ours.
• We do not report a change when two ranges overlap, even when the point numbers moved in your favour.
• We do not track prompts that contain your brand name.
• We do not publish prices for other vendors that we have not verified on a stated date. For Profound, Peec AI, Otterly.AI and Scrunch AI, go to their own sites.
• We do not claim independent benchmark results against Profound. We have not run one. If we do, we will publish the method and the date with it.

FAQs