Skip to content
George Petropoulos, home page
Article

Why an AI search can show different law-firm lists

Different AI-search answers need a clear test record. Compare questions, context, firm mentions and sources before treating a list as a stable ranking.

By George PetropoulosFounder of Inoriseo

An AI search can show different law-firm lists because the question is only part of what shapes the answer. The product, search mode, location and available context can matter too. Treat each answer as an observation from a particular test before calling it your firm’s rank or deciding that someone has manipulated the result.

If your firm appears in one answer and disappears in the next, keep both. The useful next step is to establish what was asked, where it was asked and what came back. Choosing the more flattering screenshot leaves you with weaker evidence for a spending decision.

The words you type are not the whole search

Google’s guidance on AI search features says AI Mode and AI Overviews can issue several related searches to develop an answer. It also says the two features can use different models and techniques, so their responses and links can vary.

That matters when someone compares a list from AI Mode with names in an AI Overview, the summary that can appear in Google Search. The same starting question does not make those two displays the same test.

OpenAI’s ChatGPT Search documentation describes rewriting a question into targeted searches. It says approximate location can inform local results and, when memory is enabled, relevant saved details can influence a rewritten search.

So “Which divorce lawyers are near me?” leaves a location question inside the test. Naming the city makes your request clearer, but it does not establish that every other condition matches another person’s search. A new conversation also does not, by itself, establish that saved memory is disabled.

These are reasons to record context. They do not tell you which factor caused a particular firm to appear or disappear. Avoid replacing an unexplained change with a confident story about how the system chose its names.

Keep a record that includes the missing answer

Here is a hypothetical repeated-query record. It illustrates how to describe a small test; it is not research performed on any firm or platform.

Shared setup: All example dates and times use November 9, 2026, UTC. Google AI Mode on desktop; exact question, “Which law firms handle contested divorce in Boise, Idaho?” Each check starts separately. No settings deliberately changed; inferred location and other background conditions are not established. Full answers and displayed source links are retained.

  • November 9, 10:00 a.m.: Your firm is named in the answer. Record the surrounding description and linked sources.
  • November 9, 10:10 a.m.: Your firm is named again, with a different list order and some different sources.
  • November 9, 10:20 a.m.: Your firm is not named. Keep this answer with the other two.

The fair description is that the firm appeared in two of these three illustrative answers. It does not establish a stable second-place position, a share of all AI searches or how often a prospective client will see the firm.

If you then ask, “Which of those firms offers evening appointments?”, save that as a follow-up with the earlier conversation attached. It adds a requirement and refers to a previous list. Do not quietly count it as another run of the original question.

Read the mention and the sources separately

Being named in an answer is one observation. Having your website linked as a source is another. Record each rather than treating every firm name as a citation to that firm’s site.

In the hypothetical record, the answer might name your firm while linking to a directory that discusses it. Open the actual source and check what it supports. A link beside a sentence does not settle whether every detail in that sentence is correct.

If the answer gives your old office address, you have a concrete accuracy issue to investigate. Check the firm’s public information and the cited page. You can pursue that correction without first proving why the system included your firm in this list.

A changed list alone is also insufficient evidence that a rival paid for inclusion or that your agency caused a loss. Those are separate claims requiring evidence. Keep the record of what changed and ask for the proposed explanation to be labeled as established or still being investigated.

Use repeat checks to make a bounded decision

Agree which questions matter to your practice before reviewing the answers. Keep their wording and the product or mode visible. When you intentionally change the city, service or context, label the new test so the comparison remains useful.

Ask your agency for the complete set it is discussing, including answers that omitted your firm. A summary needs to say what was tested and what remains unknown. More runs provide more observations under those conditions; they do not turn a chosen set of questions into a measure of the whole market.

Approve work against a specific finding, such as correcting an outdated public address or explaining a service your site barely describes. For an unexplained list change, agree the next check before ordering a rewrite. Keep the observed answers as the evidence, with the limits beside them.

In this article

Keep reading

Scroll to Top