What AI Says About Your Business When Chatbots Answer First

Most businesses have a reasonable sense of what their website says about them, because they wrote it. Far fewer have any idea what an AI assistant says about them, despite that answer now sitting earlier in the buying process than anything they control. The gap between those two descriptions is where a surprising amount of lost business lives.

Why this matters now

The shift is in when the summary happens. Traditionally a buyer found your website, read your positioning, and formed an impression from material you authored. Now a large share of buyers ask an assistant first, receive a synthesised description of you and two or three competitors, and arrive at your website with an impression already formed. Your carefully written homepage is now the second thing they read, not the first.

What makes this different from ordinary reputation management is the source material. An AI summary is not primarily built from your website. It is assembled from what other sources say: review platforms, business directories, industry listings, news mentions, forum threads and social profiles. Your own site contributes, but it is one voice among many, and it is the one the system trusts least because you obviously wrote it about yourself.

This creates a specific and common failure. A business rebrands, repositions, or moves upmarket, updates its website thoroughly, and does nothing about the eleven directory listings describing what it did in 2019. The AI summary reflects the eleven, not the one, because consistency across independent sources is exactly the signal these systems are built to weight heavily.

There is a second failure mode that is quieter and worse. When sources conflict, the system hedges. Instead of a confident description you get vagueness, or worse, an admission that it is unsure what the business does. In a shortlist context, being described uncertainly is functionally similar to not being described at all. The competitor with consistent information across twenty sources gets the clean summary.

The commercial exposure is largest for considered purchases with a research phase, which describes most professional and B2B services. If your buyers spend a fortnight investigating suppliers before making contact, almost all of that fortnight now happens somewhere you cannot see.

Common mistakes to avoid

The instinct is to treat this as a content problem on your own website. That instinct is mostly wrong and it wastes a lot of effort.

Rewriting your homepage to fix it. Your homepage is a weak input to these systems precisely because it is self-authored. Rewriting it again will not change a summary that is drawing on directories and reviews.
Never actually reading the output. The overwhelming majority of owners have not once asked an assistant to describe their own business. It takes ninety seconds and it is the entire diagnosis.
Ignoring dormant profiles. The directory listing you created in 2018 and forgot is still being read. Old profiles describing a former business model are among the most common causes of a wrong summary.
Chasing review volume while ignoring review content. A high star rating with reviews that say nothing specific gives a summariser nothing to work with. Reviews mentioning the actual service, sector and outcome are far more useful inputs than a numerical average.
Assuming it is static. These answers change as sources change. A quarterly check is required, not a one-off audit.

There is also a tempting shortcut worth naming. Some businesses respond by publishing pages designed purely to be quoted by AI systems, stuffed with declarative statements about their own excellence. This does not work well, for the same reason self-authored praise has never worked well, and it produces a website that reads badly to the humans who eventually arrive.

The productive framing is that you are not managing an AI. You are managing the accuracy and consistency of public information about your business, which is something worth doing regardless of who or what is reading it. Our AI search visibility playbook covers the discoverability side, and this is the reputational counterpart.

Quick Strategic Tip

Ask three different AI assistants the same question: what does [your business name] do, and who is it for. Save all three answers in a document with today's date. Do it again in three months. The differences between the three assistants tell you which sources each is weighting, and the change over time tells you whether your work is landing.

Step-by-step plan

This is a two-week audit followed by ongoing maintenance. Most of the effort is in the first pass.

Run the baseline audit. Ask at least three assistants four questions each: what does this business do, who is it for, what is it known for, and who are its main alternatives. Save every answer verbatim with the date. That last question is the uncomfortable one and usually the most useful.
Mark every factual error. Go through the saved answers and highlight anything wrong: outdated services, wrong location, wrong size, wrong sector, wrong specialism. Separate genuine errors from things you simply wish it had said.
Trace each error to a source. Search for the incorrect claim and find where it comes from. It is almost always identifiable: an old directory entry, a stale profile, an article from three years ago. This step is the whole job. Errors you cannot trace, you cannot fix.
Fix the highest-authority sources first. Not all sources carry equal weight. Your Google Business Profile, your main industry directories and your most-trafficked review platform matter far more than the twentieth listing site. Correct those before anything else.
Make your core facts identical everywhere. Business name, address, phone number, and a one-sentence description of what you do. Byte-for-byte identical across every profile you control. Inconsistency here is the single largest cause of hedged summaries.
Get specific reviews, not just positive ones. Ask satisfied clients to mention the actual service, their sector and the outcome. A review saying excellent service is nearly useless as an input. One saying they rebuilt our booking system and enquiries went up by a third is a usable fact.
Publish the things only you can state. Your process, your pricing model, your sector focus, your service area. Put these somewhere unambiguous on your own site. Your site will not dominate the summary, but it is the reference point when other sources conflict.
Earn a few independent mentions. A trade publication, a local business feature, a genuine partner page. Independent corroboration moves these systems more than anything you can publish yourself, and a small number of good mentions goes a long way.
Re-run the audit after six weeks. These systems update on their own schedules and some are considerably slower than others. Do not expect immediate change and do not conclude too early that nothing worked.
Put a quarterly recheck in the calendar. Twenty minutes, four times a year. This is maintenance, not a project, and the businesses that treat it as a project let it decay.

Step three is where most audits stall, because tracing a claim back to its source is tedious. It is also the only step that converts a complaint into an action. An error you can trace is an error you can fix in an afternoon.

Reviews deserve particular attention because they are simultaneously high-authority and something you can genuinely influence. The ask matters enormously: a client asked for a review writes something generic, while a client asked what specifically changed for them writes something a summariser can actually use. Our guide to social proof that actually converts covers how to structure that request.

AI reputation checklist

Work through this quarterly.

Three assistants have been asked to describe the business, with answers saved and dated.
Every factual error in those answers has been traced to a specific source.
Business name, address and phone number are identical across all controlled profiles.
The one-sentence description of what you do is identical everywhere.
Dormant or outdated profiles have been found, updated or removed.
Recent reviews mention specific services, sectors and outcomes.
Your own site states your process, pricing model and service area unambiguously.
At least one independent third-party mention exists from the past year.
The alternatives question has been asked and the answer noted.
A recheck date is in the calendar.

How to measure impact

This resists conventional analytics, because the impression forms somewhere you have no tracking. Measure it qualitatively and accept that the instrumentation is imperfect.

Keep a dated log of the summaries. The primary measurement. Same questions, same assistants, every quarter, saved verbatim. Improvement here is the outcome, and without a log you will have no idea whether anything changed.

Count factual errors over time. A simple number that should fall. Six errors in January and one in July is unambiguous progress you can report.

Track how often you appear in alternatives answers. Ask an assistant to recommend suppliers for the service you provide, in your area. Whether you appear, and how you are characterised, is the closest thing to a ranking in this context.

Watch branded search volume. If more people are searching your name directly, something upstream is putting you in front of them. Combined with flat or falling non-branded clicks, that pattern suggests AI-mediated discovery is working.

Ask new enquirers how they found you. The most direct evidence available. Answers referencing an AI assistant have gone from nonexistent to routine over the past two years, and they are worth capturing explicitly.

Key terms in plain English

AI summary: The description an assistant generates about your business, synthesised from multiple public sources rather than any single page.

Source authority: How much weight a system gives a particular source. Established directories and review platforms outrank obscure listing sites.

Consistency signal: Agreement about your business facts across independent sources. The strongest lever you have and the most commonly neglected.

Hedging: When conflicting sources cause a system to give a vague or uncertain answer. Commercially close to being invisible.

Independent mention: A reference to your business on a site you do not control and did not pay for. Weighted far more heavily than self-published material.

Dormant profile: An old listing you created and forgot. Still readable, still influential, frequently wrong.

Conclusion and next move

You cannot edit what an AI says about your business, and the instinct to try is what makes this feel hopeless. You can change what it reads, and that turns out to be ordinary work: accurate listings, consistent details, specific reviews, a few independent mentions. Unglamorous, tractable, and largely neglected by your competitors.

Spend ninety seconds today asking an assistant what your business does. If the answer is wrong, outdated or vague, you have found something worth fixing that is currently shaping decisions before anyone reaches your website. Most owners have never looked, which is exactly why the advantage is still available.