Deep Dive inside the HiBot AI Visibility Audit
5 Oct
I have now run and published 10 AI visibility audits on brands everyone knows, one a week, start to finish, in public. Every time I post one, somebody asks the same thing: what is actually in the report? Fair question. The headline is one number on the first screen, and if that is all you ever see, you are missing the part that does the work. So here is the whole thing, opened up, component by component, using my Toyota audit as the running example.
First, thirty seconds on why anyone should care.
You Are Being Judged in a Room You Cannot Enter
A few years ago you could watch a customer decide. Search, a few tabs, a review site, then your site, and you saw all of it in Analytics. That path has collapsed into one AI conversation that happens before the customer ever lands on you. Whichever brand the engine names in that single answer usually wins, and most companies have never read one word of what it says about them. I call that shift the customer journey singularity, and it is already here.
The numbers are not gentle. Gartner reported in 2026 that 72% of consumers now bump into generative AI across their everyday internet use whether they asked for it or not. They do not fully trust it: in the same study, 54% of the people who used AI to shop ended up double-checking everything it told them, and 62% felt it wasted their time. Then Semrush surveyed more than 2,300 US adults in July 2026 and found 57.5% of AI users have talked themselves out of a purchase because of something a chatbot said.
Here is the part that should bother you. AI reaches most of your customers uninvited, it is wrong often enough that they have learned to second-guess it, and it is already closing the door on more than half of them before you get a word in. You cannot fix a conversation you have never read.
The cheapest way to start reading it is free. Enter your website at hibot.com and in about 30 seconds you get the exact questions your customers are asking AI about you, no account needed. That is the front door. The audit is the house. Let me walk you through the house.
A quick note before we go in: the 10 brands I have audited are all public, top to bottom, which is the closest thing there is to seeing what you are buying before you buy it. Oura Ring, Costco, Accenture, Salesforce, American Express, Delta Air Lines, Marriott, FedEx, Nike, and Toyota. Read any of them alongside this.
The Scorecard
Everything opens on one screen. The Score Dashboard gives you the headline AI Visibility Score from 0 to 100 with its band from Invisible to Dominant, the 6 ANSWER dimensions with the weakest one flagged, and two rates that tell you how the engines are behaving: how often they cite a source about you, and how often they get the facts right. Toyota reads 93 and Dominant, Reputation perfect at 100, Nomination flagged as the soft spot at 83. Accuracy 99%, citation rate 11%. Run the audit again later and every number carries a delta against the last one, so the screen becomes a before-and-after instead of a snapshot.
Right under it, Competitive Standing says the quiet part out loud: how often AI names you, how often it names the category leader, and where you rank. Toyota is named in 53% of answers and ranks first of four against Honda, Ford, and Tesla. Coverage by Model then splits that by engine, because a brand can own one engine and vanish in another. Toyota shows up in 96% of ChatGPT’s answers and 100% of Perplexity’s and Google’s. And Share of Voice turns every mention into a ranked leaderboard with a first-mention breakdown, so you catch a rival quietly winning the recommendation in answers where you are technically present. Being mentioned and being chosen are not the same thing, and this is where the gap shows.
Watching It over Time
One audit is a photo. A recurring one is the film, and since AI answers jump around week to week, the film is what you actually want. Put a brand on a schedule and every report picks up an over-time layer. Dimension Trends plots all 6 ANSWER categories across every run and flags what moved and what stalled. Question-Level Changes goes one question at a time to show where you newly appeared, where you dropped out, and where a competitor passed you, which is usually the most convincing single page an agency can put in front of a client. Accuracy Over Time tags each flagged claim as new, persisted, corrected, or reappeared, so you can prove a bad claim is finally dead. And Timeline Notes let you pin the changes you made, with dates, against the score line, with the honest caveat that nearby movement is a correlation, not a trophy.
The Action Plan
Now the report stops describing the problem and starts handing you the work. The Impact-Effort Prioritization Matrix plots every recommendation by impact against effort and sorts the pile into quick wins, big bets, fill-ins, and time sinks. It is the fastest way to decide what to do this week and what to leave alone, and it recolors as you check things off.
The ANSWER Deep Dive is a plain-language read on each of the 6 dimensions, with the exact customer questions behind them and a one-click path to the word-for-word answers, so nothing is a black box. This is where Toyota’s real story sits. Reputation is perfect, Showdown is nearly perfect, but Nomination, the dimension that measures the actual recommendation, is the weak one at 83. Asked to pick a vehicle for a specific family, ChatGPT led with the Subaru Ascent and never mentioned Toyota. Perplexity walked through the Sienna, the Highlander, and the Grand Highlander, then told the reader to go buy the Kia Sorento Hybrid. Toyota wins the brand question and loses the scenario question, and the Deep Dive shows you the exact prompts where it happens. The Tracked Strategic Recommendations turn that into a to-do list: each fix ranked by priority, backed by a real quote from the audit, with the problem, the evidence, the steps, and the expected score lift, plus a status and notes you set right on the report. It is a working tracker, not a slide that dies in a deck.
The 90-Day Roadmap sequences those fixes into fixed horizons, puts a score target on each, and drops every milestone into your calendar or project tool in one click. Dates, owners, and targets are the difference between a report that gets worked and one that gets admired and filed.
The Evidence
This is the part that makes everything above it believable. Every score traces back to a sentence an engine actually produced.
Claims AI States as Fact collects every flagged claim an engine made about you, checks it against a snapshot of your own site, and sorts it by origin: a model error the AI invented, your own words repeated back as fact, or a figure it lifted from a third party. Each one needs a different fix, so the log tells you which is which, and you get the final call on every row. In Toyota’s audit, Google capped 3 otherwise favorable answers on invented detail: an unsupported market-share figure, a claim that Toyota Safety Sense comes standard on virtually all new Toyota vehicles, and a wrong definition of the LE trim. Good answers, quietly undercut by a fabricated fact, now caught and labeled.
Where You Show Up is a grid of every question against every engine, each cell showing whether you were named, absent, or got nothing back. Click a cell to read the real answer, and on a re-audit a compare mode overlays what flipped. It is the whole audit at a glance.
The Citations and Sources Log lists every source the engines leaned on, de-duplicated, with the engines that cited each. This is where Toyota’s one weakness traces to its root. The audit captured 196 sources: 194 third-party, just 2 of Toyota’s own, and none from a competitor. An 11% citation rate means the engines are talking about Toyota without reading Toyota, which is exactly how an invented market-share figure slips through. The log hands you the specific domains to go earn a mention on, or to correct.
And the Verbatim Responses Explorer holds the full, word-for-word answer behind every score, captured by hand in a neutral session, filterable by category and engine, each with its score and the reason for it. You read exactly what your customer would have read. Nothing is paraphrased, so nothing rests on trust.
In Your Customer’s Language
If your customers ask AI in German or Slovak, an English-only audit is measuring the wrong market. Set the customer language once and every question is generated natively, phrased the way locals actually type, then asked and captured in that language from the first prompt to the last, across 26 languages. Geography and language are locked into each run, so your trends still compare like for like.
Take the Report Anywhere
A report only earns its keep if it can leave the browser. Export the whole thing as an editable Word document, a print-ready PDF, or a CSV of every question and answer for your own analysis. If you are an agency or a consultant running this for clients, you can white-label the entire deliverable, your logo and report name on every page, every HiBot reference neutralized to a clean “AI Visibility Score.” And over MCP, the AI assistant your team already uses can read the delivered audit directly, its scores, verbatim answers, claims log, and trends, and even mark recommendations done as the work gets finished.
Why This Is the Work That Matters Now
Toyota is the reassuring version of the lesson. A brand can post a 93, lead its category, and still lose the sale the second a customer asks which model to actually buy, and still carry a fabricated fact capping its best answers. Most brands are not starting from 93. The point of the report is not the number on the front. It is that every score traces to a specific answer, every answer traces to a source, and every problem shows up with a fix, a priority, and a date. That is the whole distance between knowing AI matters and doing something about it.
A Few Questions I Get
Do I need the audit to get value here?
No. Generate your free question set first, read it, and you will already know more than most of your competitors. The audit is for when you want the answers captured and scored properly instead of guessing.
Why Toyota as the example?
Because it is the cleanest version of the pattern every one of the 10 audits found: AI knows the brand and still hands the recommendation to someone else. If it happens to Toyota at a 93, it is happening to you.
Can I see a full report before I buy?
Yes, that is the point of publishing all 10. Start with Costco or Toyota and read one end to end.
The Take
Every market now is one where customers ask AI to compare their options, which means a report you have never read is out there deciding your deals. You can see the first page of it free in about 30 seconds, and you can read 10 complete ones on brands you already know. There is no reason to be the last person in your market to find out what the machines are saying about you.
See the questions your customers ask AI, free at hibot.com. The full feature breakdown is at hibot.com/features, and a new audit goes up every week on my LinkedIn.
Every HiBot audit is run by hand by a human AI specialist and scored independently on the ANSWER methodology, our approach to answer engine optimization (AEO) and generative engine optimization (GEO). When an engine invents a product or a promise, we report it as the engine’s output, disclose that the audit is independent, and keep the net story fair to the brand. Grab the ANSWER whitepaper for the full framework, or request an audit at hibot.com.
