Key takeaways
- Some of it is visible as referral traffic from assistant domains, and that portion is easy to segment.
- Most of the influence is invisible, because the user reads an answer and then searches for you directly.
- Direct and branded search lift is the practical proxy, imperfect but directional.
- Ask on the form. A how-did-you-hear field remains the most useful instrument available for this.
- Report it as a confidence range, not a number, or it will be either dismissed or over-trusted.
Measuring AI assistant influence is genuinely hard, and the honest starting position is that a substantial part of it cannot currently be attributed. That has not stopped confident numbers circulating in both directions, and a franchise marketing team is better served by an accurate picture of the uncertainty than by a precise-looking figure.
What follows is what can be measured with ordinary tooling, what cannot, and how to report the combination without overclaiming. This is the measurement companion to auditing what AI assistants say about your brand.
The part that is visible
When a user clicks a link inside an assistant's answer, that visit usually arrives as referral traffic from the assistant's domain, and it can be segmented like any other referral source.
- Build a referral segment covering the assistant domains you care about, and review which ones actually appear rather than assuming
- Expect low volume and high engagement. These visitors have generally read an answer about you before arriving, and behave more like a referral from a trusted source than like search traffic
- Segment by landing page. Which pages assistants send people to tells you which of your content is being cited, which is otherwise hard to learn
- Watch it as a trend, not a total. The absolute number understates influence badly; the direction is more informative
Separately, server logs and crawler analytics will show assistant-operated crawlers fetching your pages. That is an indicator of retrieval interest rather than of user demand, and it is worth watching for a different reason: it tells you whether your pages are being fetched at all, which is the precondition for everything else.
The part that is not visible
The larger share of assistant influence produces no referral at all. A user asks a question, reads an answer that names your brand, and then does what people have always done with a recommendation: searches for the brand directly, or types the name into a maps application, or calls the location.
From an analytics standpoint that visit is direct or branded organic traffic, indistinguishable from someone who heard the name from a neighbour. No amount of tag configuration recovers the original cause.
| What happened | How it appears | Recoverable? |
|---|---|---|
| Clicked a link in an answer | Referral from the assistant domain | Yes |
| Read the answer, searched the brand | Branded organic or direct | No, only inferable |
| Read the answer, called the location | Phone call, no digital trail | Only by asking |
| Read the answer, went to a maps app | Local profile interaction | Partially, via profile insights |
| Assistant summarised without naming a source | Nothing at all | No |
The proxies worth watching
Since the direct measurement is incomplete, the practical approach is a small set of indicators that move together when assistant visibility improves.
- Branded search volume, watched for lift that does not correspond to a campaign or a season
- Direct traffic to deep pages. Direct traffic to the homepage is ambiguous; direct traffic to a specific location or service page usually means someone was told about it
- Assistant referral trend, as above, as the visible tip of the distribution
- Citation presence over time, tracked by periodically asking the assistants the questions that matter and recording whether you appear
- How-did-you-hear responses, which remain the single most useful instrument and are usually the only one that catches the phone path
- Local profile interaction data, which can show demand arriving without a website visit at all
The fourth item is the one most worth building into a routine, because it measures the thing itself rather than a downstream effect. A fixed list of questions, run on a schedule, with the answers recorded, turns an unmeasurable channel into a trackable presence metric.
Adding the question to your forms
A how-did-you-hear-about-us field is imprecise, self-reported and unfashionable, and it is currently the best instrument available for this particular problem. It is also the only one that works for phone enquiries, which in most franchise verticals is where a large share of the demand arrives.
The practical version: an optional free-text or short-list field, with an option that names assistants explicitly rather than hoping people write it in. Enough respondents mention it unprompted that having the option available produces a usable signal within a quarter.
Reporting it without overclaiming
The temptation is to produce a single number. Resist it — the number will be wrong in a direction nobody can quantify, and it will be used to justify either abandoning the work or over-investing in it.
A more defensible report states three things: the measured referral volume, which is a floor rather than a total; the proxy indicators and their direction; and the citation presence, which is the actual objective. Presented that way, the picture is honest, comparable over time, and immune to the accusation that follows any precise-looking figure in this area — which matters more than usual given how much is currently being asserted with more confidence than the data supports.
Questions we get asked
Can traffic from AI assistants be tracked in analytics?
Partially. Clicks on links inside an answer usually arrive as referral traffic from the assistant's domain and can be segmented like any other source. The larger share of influence produces no referral at all, because the user reads the answer and then searches for the brand directly, which appears as direct or branded organic traffic.
What is the best proxy for AI assistant influence?
A small set of indicators watched together: branded search lift with no campaign or seasonal explanation, direct traffic to deep pages rather than the homepage, the visible assistant referral trend, and citation presence tracked by periodically asking the assistants the questions that matter.
Is a how-did-you-hear-about-us field still useful?
For this problem it is currently the best instrument available, and it is the only one that captures phone enquiries — which in most franchise verticals carry a large share of demand. Include an option that names assistants explicitly rather than relying on people to write it in.
How should AI assistant performance be reported to leadership?
As a range with stated limits, not a single number. Report measured referral volume as a floor rather than a total, the proxy indicators and their direction, and citation presence as the real objective. A precise-looking figure in this area will be wrong by an unquantifiable amount and used to justify the wrong decision.
What does assistant crawler activity in server logs tell you?
That your pages are being fetched, which is the precondition for being cited but not evidence of user demand. It is worth watching as a technical health signal — particularly if it stops, or if it never covered a section of the site in the first place.
Want this mapped to your system?
We will walk your recruitment funnel and your franchisee marketing on one call, and show you where it is leaking. No obligation.
Book a Demo