Law firm marketing and AI: review what you publish, record the rest

Who reads a page before it goes live?

The responsible attorney reads it: the lawyer whose practice the page describes, or the attorney the firm has named as responsible for what it publishes. Where the firm has ethics counsel, in-house or outside, counsel reads it too. Whether the rules where you practice name a person who must approve advertising is a question for your bar; the routine needs a name in the record either way. This page describes a routine, not the rules it serves, and it is not legal advice.

The reading covers everything the firm writes about itself, wherever it appears: practice area pages, attorney pages, blog posts, the questions page, and the text the firm writes on other sites, such as the Business Profile description and posts, the directory profiles it controls and its social posts. A sentence an answer repeats can come from any of them, and the firm wrote all of them. What the pages themselves should state is covered in the overview for law firms; this page is about the reading.

Whoever drafts the page is not the one who approves it. Marketing staff, a web developer or an agency, this one included, can write a draft and press publish; the approval that sends it live belongs to the attorney. Keeping the two roles apart is what gives the record its meaning.

What the reader holds the page against

Four references, each kept where the reader can open it while reading. The page is checked against each in turn, and the record later names the version of each that was used.

The word list is the firm's own. A firm that took the questions on the page about facts and conduct rules to its bar writes the answers here: each word, what was decided, who decided and on what date. The list grows as answers come back, and each new entry is a reason to reread every page that uses the word.

The reader reads the markup too. The JSON-LD on a page is published by the firm like any sentence on it, though no visitor sees it. Google's structured data introduction says it should describe only what is visible, and its general guidelines ask for up-to-date information and nothing misleading. A reader with no use for code can ask the developer for the structured data as a plain list of the statements it makes.

The approval record, and the two dates on it

Each reading ends in a record, one per version of a page. It names the page, what changed, who read it, what it was checked against and what was decided, with two dates: the day of the approval and the day the page went live. The two dates can differ, and anything edited between them is a new version that needs its own reading.

page:             [address, or the profile and the platform]
version:          [one line on what changed]
drafted by:       [name or role]
read by:          [responsible attorney]; [ethics counsel, where the firm has one]
checked against:  rules for [jurisdiction], as of [date]
                  attorney pages, register checked [date]
                  word list, version [n]
                  intake terms as of [date]
decision:         [approved | returned, with the reason]
approved on:      [YYYY-MM-DD]
live on:          [YYYY-MM-DD]
copy kept:        [file of the page exactly as published, markup included]
read again:       [a date, or the change that will trigger it]

Keep the copy of the page as it went live beside its record. The page will change, and someone may later need to know what it said on a given date, markup included. How long such records have to be kept is a question for your bar; until you have the answer, keep all of them.

When the reading happens again

A page that was right when it was approved can become wrong without anyone editing it, because the firm changed around it. So a reading is triggered by changes at the firm as well as by changes to the page:

• an edit to any sentence that states a fact about the firm or makes a claim, however small the edit;
• an attorney joining, leaving, being admitted somewhere new or changing name;
• a change to the consultation terms or to the fee basis for any matter;
• an office opening, moving or closing, or changing its hours;
• a change to the rules where the firm is admitted, or a new answer from counsel that adds to the word list;
• an edit to a profile on another site, whether the firm made it or the platform did;
• a fixed date once a year, for a full reading of every page and profile.

When a trigger fires, the re-reading follows the fact rather than the page:

1. Read the page where the fact changed, and record it as a new version.
2. Search the site for the old wording, and read every page that still repeats it.
3. Read the markup on each of those pages.
4. Update and read the profiles on other sites that state the same fact.
5. Record each one with its own dates.

Answers the firm does not control

An AI answer is written for one person and published nowhere the firm can edit it, so it never passes through the reading. When an answer describes the firm with a claim the firm would not make, such as "the top DUI firm in [county]", "[attorney], a certified specialist" or a result the firm never announced, the firm cannot change the answer. It can find where the claim is written and correct what it controls:

1. Record the answer whole: the question, the assistant, the date, the claim in the answer's own words, and any page it cited.
2. Search the firm's own pages, markup included, for the words. If they are there, the firm said it, and the page goes back through the reading above.
3. Check the profiles elsewhere: the Business Profile, directory and rating-site profiles, each attorney's own profiles, old press releases and guest posts. Each is a brand mention, and a badge a site awarded, a category a directory assigned or a years-old headline can carry a word the firm would not use.
4. Correct what the firm controls. For what it does not control, ask the platform in writing, and note the date and the reply.
5. Keep the record with the routine's other records, and ask the same question again a month later. A firm with a monthly panel already has a column for claims like this.

The general procedure for wrong statements about a business, including how to tell a bad source from a hallucination, is in when AI gets facts about your business wrong. What a law firm adds is the record: it shows what the firm published and when, and what it did once it found a claim it had not made. Whether the firm owes anything more over a claim a third party made is a question for its ethics counsel.

An answer written from a model's training data can repeat an old claim after every page has been corrected, because that data stops at a knowledge cutoff. The dates in the record are what let the firm show when each correction was made.