Digital Marketing Expert Witness · Reference Article

AI Search Visibility Disputes: New Surface, Old Evidentiary Questions

Generated answers are non-deterministic, personalised and unarchived. That makes capture discipline the whole ballgame.

What is actually new

Search used to return a list of documents. Increasingly it returns a composed answer that summarises, paraphrases and selectively cites those documents - and a growing share of users never reach the underlying page at all.

That shift produces genuinely new claims: that a system misattributed a statement to a business, that it reproduced content without meaningful citation, that it stated something false about a company, or that a competitor's material was surfaced in preference to the claimant's. The subject matter is new. The evidentiary discipline required is not, and an ai search expert witness is largely applying old rigour to an unfamiliar surface.

Three properties that break naive evidence

  1. Non-determinism. The same query can produce materially different answers minutes apart. A single screenshot establishes that an output occurred once, not that it is the system's behaviour.
  2. Personalisation and context. Account state, location, history, interface version and session context all influence output. An answer reproduced without that context is not comparable to the one complained of.
  3. No archive. Unlike a web page, a generated answer leaves no public historical record. If it was not captured at the time, it is frequently gone.

The practical consequence is that these matters are usually won or lost on preservation decisions made before anyone thought there was a case.

Capture protocol as the core deliverable

Where an expert adds the most value early is in specifying how to capture, not in interpreting what was captured badly. A defensible protocol records, together:

Repetition matters most. A claim that a system "says" something is a claim about distribution of outputs, and that requires trials, not an anecdote.

The commercial-impact question

Claims that AI systems diverted business face the same causation discipline as any other visibility matter. Referral data from AI surfaces is incomplete and inconsistently attributed, branded search demand shifts for many reasons, and the whole category has been changing rapidly enough that year-over-year baselines are unreliable.

An honest analysis separates measurable referral change from inferred substitution, and declines to convert the latter into a number the data cannot support. That restraint is also what makes the measurable part credible.

Scope discipline again

Most of these disputes also contain a genuinely technical question - about training data, model behaviour, retrieval architecture or output filtering - that belongs to a different specialist. The ai expert witness practice at Stratex Digital Marketing is explicitly scoped to applied and commercial use: how organisations deployed AI, what oversight they exercised, how AI-assisted content reached consumers, and what marketing evidence shows about impact. Where source-code, architecture, forensic or advanced statistical work is required, a separately qualified specialist is coordinated in rather than the scope being stretched.

Dan Stratford has worked with generative AI tooling in commercial marketing since 2023, which is the applied grounding for that scope. Case enquiries: 720-985-7945, or see the ai expert witness Dan Stratford page and his independent ai expert witness directory listing.

Related expertise

Frequently asked questions

How do you preserve an AI-generated answer as evidence?

Capture it contemporaneously with the full query, timestamp, interface, account state, location and complete response including citations, and repeat the query across multiple trials to establish whether the output is characteristic.

Why can't an AI answer simply be reproduced later?

These systems are non-deterministic and personalised, and outputs are not publicly archived. Model and interface updates also change behaviour, so an answer produced weeks later is not evidence of what was produced before.

Can loss of traffic to AI search be quantified?

Partially. Referral attribution from AI surfaces is incomplete, so a defensible analysis separates measurable referral change from inferred substitution and states the limits of what the data supports.

Does an applied AI expert review source code or model architecture?

No. Those belong to separately qualified technical specialists. An applied expert addresses selection, supervision, disclosure, measurement and commercial reliance, and coordinates with a technical expert when the facts require one.