# Chapter 4 · Which clients and which questions to go after first > GEO Playbook · Law Firms v1.0 · Canlah AI · CC BY 4.0 · Web page: https://canlah.ai/playbook/law/pick-questions/ > Markdown edition for AI assistants, same content as the web page. Figures are code blocks (wireframe / mermaid / bars / steps / split); "→" links open the matching section on the web, and the same URL with .md is its Markdown edition. **Read the General Edition first**: （→ 通用版 4.5 买家归堆、年机会值与三道关）, （→ 通用版 4.6 四态、两层分母与弃权线）, （→ 通用版 4.7 逐态处置、十类问题与本季名额）. This chapter covers only the three places where law firms differ from the General Edition: the unit for grouping is one type of matter; the "not-reachable" list within the four states, and the two thresholds that hold law firms back; and where each type of question goes — fees, near me and "is this firm good" come first. ## 4.1 Group by matter type and work out what each group brings in a year **What you'll do in this section**: take out the firm's fee records and its lawyer time records, group them by "one type of matter", work out an annual opportunity value for each group, and rank them by that number to decide which group to go after first. Get the monthly capacity cap from the time records; don't estimate it in a meeting. The annual opportunity value is for ranking only: it goes into no external material, and into no quote we give the law firm. ```bars Figure: An example: one family-law firm's books, for ranking only, not a revenue forecast; the numbers are illustrative values built from the formula unit: S$ Contested divorce | 288000 | capacity capped by lawyer hours Uncontested divorce (fixed fee) | 180000 | the fees page's main battlefield Probate | 120000 | close to the next group; first compare which is reachable enough Wills + LPAs (Lasting Powers of Attorney) | 115200 | competitors aren't only law firms Prenuptial agreements | 60000 | small volume, ranked last ``` The work unit is **one type of matter**: practice line × client type, such as "contested divorce · individual", "shareholders' agreement · SME owner", "wills + LPAs · older couples". Group by how the firm opens matters and charges for them on its own books, not by how questions are asked — the same client will ask about the process first, then fees, then near me, but it's the same piece of business being asked about. The five numbers in the figure aren't any real firm's books, and they aren't market rates — swap in your own. The formula and the rules follow （→ 通用版 4.5 买家归堆、年机会值与三道关）: **annual opportunity value = price per order × gross margin × monthly capacity cap × 12**. Law firms fill in these four numbers with two differences from other trades: - **Price per order** is the firm's own legal fees actually collected on the most recent batch of closed matters, excluding court fees and disbursements paid on the client's behalf — that money is paid to the court and third parties on the client's behalf; it isn't the firm's own revenue. - **The monthly capacity cap is calculated from lawyer hours**: lawyer hours that can go into this type of matter each month ÷ the average lawyer hours a matter takes from intake to close = the most new matters that can be taken on each month. Both numbers are copied straight from the time records. A contested divorce can drag on for months; calculate by "how many new ones can be taken on each month", not by how many are currently open. > **Example** (illustrative values, not market rates): the contested-divorce cell in the figure: price per order S$20,000 × gross margin 40% × monthly capacity cap 3 matters × 12 = S$288,000. Copy the rules down together with the numbers. The first four follow the General Edition; the fifth is added for law firms: 1. Fill in the numbers yourself, confirm them in person, and put them somewhere visible in the plan. 2. **The monthly capacity cap is a hard cap** — if the time records say you can only take on 3 contested divorces a month, don't pick a target calculated on 8. 3. The annual opportunity value is for **ranking only**; not one word of it may be presented externally as a revenue forecast. Never multiply, divide or display it side by side with seat counts or enquiry counts. 4. When two groups have close annual opportunity values, pick **the group with less competition** — probate and wills + LPAs in the figure are a case like this; use 4.2 first to count which group has more reachable-enough positions. 5. **The annual opportunity value goes into neither our quote nor our contract.** We only charge law firms a fixed service fee — never per lead, per case, per signed client, nor as a percentage of legal fees or the matter's value: charging that way could make us guilty of an offence under s 33(3) of the Legal Profession Act (LPA), and that criminal liability is our own; on the law firm's side, it runs into PCR r 19 (not sharing legal fees with a non-lawyer, not paying commission) and r 39(2)(b) (no commission or any other consideration for a referral) [Statute text; see Appendix A.1, A.2]. The fixed service fee itself [Original text not obtained; see Appendix A.6]: no prohibition found in the current rules (scope checked: all of LPA s 33; check date: 2026-09-29), 70% confidence. **For the wills + LPAs and probate groups, work out whether they're reachable enough before ranking them** — on the ChatGPT leg, these two groups are cited almost entirely by government, and the competition isn't only other law firms: - In the family set (public dataset R1, 2026-09-23, each question run once), for the question on LPAs, all 5 ChatGPT citations were government pages; for "how much does it cost to write a will", 4 of 7 were government pages and the SAL Wills Registry, and the other 3 were two will-writing companies and a wills directory site that calls itself not a law firm (entities labelled by domain and title, not checked page by page) — 0 law firms. On the LPA question, AI Mode also cited an article from a healthcare group about LPAs (single case) — doctors who can sign LPA certificates are in this group too. - L3 (the 2026-09-29 run, detailed below) has 4 questions in the wills, estates and LPAs practice area (3 on probate, 1 on how an estate is divided when there's no will): in ChatGPT run 1, 16 of 17 citations were government and public-body pages and 1 was a law-firm website, on the question about probate legal fees (in run 2, too, the only law-firm citation, 1, was on this question); AI Mode cited a law-firm website on all 4 questions (11/20). Conclusion: on the ChatGPT leg, so far the only way into these two groups we've seen is through fee questions; explainer pages feed the AI Mode leg only. The two legs have different denominators — read only who's cited, don't compare the sizes. Law firms are a professional-services buyer type. The General Edition's cross-industry research says this type wins through educational content (the sample doesn't include Singapore, 75% confidence; evidence in （→ 通用版 A.2 选点、写页、站外与复测的证据）), and the tactic is to add long-form content and get onto directories and lists. Law firms should play it the same way, but remember that in this trade, long-form content feeds the AI Mode leg only (numbers in 4.2); the ChatGPT leg depends on fee pages, office pages and directory listings. Reviews are not the main lever; for how to ask for them and what conditions apply, see chapter 6, 6.5. Groups like contested divorce and criminal defence — where the client calls first, books a consultation first, and signs the letter of engagement later — do not drop out of the pool just because they're hard to attribute: attribution is a site-level capability, not a property of the question. On the day you start, put it on the "observable tier" for acceptance; for how the two tiers are split, see the three checks in （→ 通用版 4.5 买家归堆、年机会值与三道关）. Add a required "How did you find us?" field to every form, phone call and WhatsApp chat (chapter 2, 2.1). ## 4.2 Where law-firm questions should land: fees, "near me" and "is this firm good" come first **What you'll do in this section**: first work out exactly which question types get ChatGPT to cite law-firm websites; judge the four states for every URL in the intent-level top 10, and fix the law firm "not-reachable" list as written below; then assign an action by question type — the first batch takes only fee questions, near-me questions and brand questions like "is this firm good"; process, definition and rules questions don't go in the first batch, and whatever's left is accepted on the AI Mode leg only; policy questions like legal aid get rewritten as execution questions before being judged again; finally, break each priority question down into the things AI will check, and match each one to the page or off-site entry you need. ```bars Figure: ChatGPT cites law-firm websites only in four question types (L3 run 1) unit: citations Price | 5 | /23 · 3 of 5 questions Near me | 3 | /8 · 2 of 2 questions "Is this firm good" | 3 | /11 · 1 of 2 questions "Who's best" | 1 | /12 · 1 of 3 questions Process | 0 | /43 · 7 questions Rules | 0 | /20 · 5 questions Checklist | 0 | /13 · 3 questions Definition | 0 | /10 · 2 questions Calculation | 0 | /6 · 1 question ``` L3 is the 2026-09-29 run: 30 questions outside family law (26 English, 4 Chinese), covering 8 practice areas and including 5 questions on choosing a lawyer and legal fees that come up across every practice area; ChatGPT was asked each question independently twice, AI Mode and the Google top 10 once each; the figure plots ChatGPT run 1. Four things the figure can't show: - **Run 2 is largely the same**: law-firm websites appear only in price 6/26, "is this firm good" 2/10, "who's best" 2/14, checklist 1/19. Near-me is 0/8 in run 2 — but both answers still listed law firms' names and addresses, with the source swapped for third-party pages. Being named in a near-me answer doesn't have to come from your own page being cited (see the paragraph on near-me questions below). - **The family set's 22 questions (2026-09-23, each run once) give the same conclusion**: ChatGPT cites law firms only for price (3 of 6 questions), near me (1/1) and "is this firm good" (1/1); the "who's best" question is 0/1, and the remaining 13 questions — process, requirements, definition, rules, statistics, calculation, checklist, complete guide — are 0/13. - **The AI Mode leg runs the opposite way**: at least one law-firm website was cited in 24 of L3's 30 questions, including 6 of 7 process questions and 4 of 5 rules questions. - **Small samples**: each type has only 1–7 questions, calculation has just 1, so the proportions are directional only. ### The four states: the not-reachable list and the two thresholds that hold law firms back ```mermaid id=four-states Figure: how a URL is assigned one of the four states; "to ask" is only a transitional state and must settle into "reachable" or "not reachable" before the baseline flowchart LR q1{"Is your name on the page?"} -->|yes| a["Already present"]:::hl q1 -->|no| q2{"Is it one of the five not-reachable classes?"} q2 -->|yes| d["Not reachable"] q2 -->|no| q3{"Can you point to an entry point?"} q3 -->|yes, can point to it| b["Reachable"]:::hl q3 -->|can't tell, and it hasn't said it's closed| c["To ask"]:::warn c -->|reply to the enquiry says open| b c -->|no reply after 7 days| d ``` How the four states are judged, the four thresholds for "reachable enough", and the abstain line are fully defined in （→ 通用版 4.6 四态、两层分母与弃权线）. What law firms need to fill in are the specific values in the cells below. **Not reachable, judged by category directly** — pages belonging to government, the courts and statutory bodies, and the registers they maintain: - The courts: judiciary.gov.sg (including the Family Justice Courts), eLitigation judgment pages; legislation: sso.agc.gov.sg. - The Ministry of Law, mlaw.gov.sg, including the Legal Aid Bureau and the Public Defender's Office. - The Ministry of Manpower and the Tripartite Alliance for Dispute Management (TADM), ACRA, CPF, HDB, MSF (including its LPA pages), and government Q&As on ask.gov.sg. - The Law Society of Singapore's Practice Direction pages and its fee-dispute page. - The LSRA register lookup for lawyers and law practices, and the SAL Directory of Specialists — check only your own row; don't submit content to them. **Competitors' own domains are not reachable either**: other law firms' websites, including any keyword microsite they run under a practice-area name (the kind whose footer says the firm operates it). This is the category cited most on the AI Mode leg — 47 of L3 AI Mode's 122 citations were law-firm websites, and all of them are not reachable except your own; the only way to compete for position on this leg is to write your own pages. **Judge by URL, not by domain**: under the same Law Society domain, Practice Direction pages and the fee-dispute page are not reachable, but a law firm's listing page in the directory is reachable — the Law Society's own text states "the listings are paid by advertising" [Statute text; see Appendix A.5], but whether every listing is a paid placement, and the rate card, were not obtained this round. **The three most common "already present" cases for law firms**: the firm's own entry in the Law Society directory, an individual lawyer's tier in a ranking guide, and the firm's entry in a rating directory. The first thing to do is make the practice areas, office and fee basis on the entry match the firm's website word for word; for how to send a correction when an entry is wrong, see chapter 3, 3.4. **Of the four thresholds for "reachable enough", law firms get stuck on two**: - **"Allowed on this side"**: no platform that refers, assigns or matches cases for law firms counts as reachable enough, whatever it charges — per lead, per completed engagement, commission or a fixed entry fee. Charging per lead, per completed engagement or commission [Statute text; see Appendix A.1–A.3]; charging a fixed entry fee [Conservative line (not statute text)] (80% confidence). A "best" list spot you can only buy your way onto, or an award you get simply by paying, also doesn't count [Conservative line (not statute text)] (70% confidence). A "best" list a third-party editor writes themselves, with no payment: a law firm appearing in a third party's publicity is governed by r 46, and once in, it still has to follow r 43 and r 44 [Statute text; see Appendix A.1]; for lists like this we only do two things — supply checkable facts and send correction letters — never pay, never ask for a ranking [Conservative line (not statute text)]. - **"Free, or the rate card is public"**: the Law Society's own "Find A Featured Lawyer/Law Firm" directory (the Featured directory), which it sells, is the only paid placement worth considering for law firms [Conservative line (not statute text)] (70% confidence) — listed by practice area, chosen by the reader, with no referral or case assignment. Buying this kind of pure advertising slot at a fixed rate-card price [Original text not obtained; see Appendix A.6]: no prohibition found in the current rules (scope checked: PCR r 39, GN 6.1.1 paras 33–37, PD 6.1.2, the Law Society's Featured directory page; check date: 2026-09-29), 70% confidence. But its rate-card terms were not obtained this round (the original page returns a 404) — **until you get the rate card, this one doesn't count as reachable enough.** The full method for judging paid placements is in chapter 6, 6.2. Judge ranking guides (Doyle's, Legal 500, Chambers, Best Lawyers and the like) URL by URL: one with an open submission channel counts as reachable; one that only takes peer nominations and doesn't accept self-nomination from firms counts as not reachable, as a "closed-door rating". This round didn't check each one's submission process and whether it charges; ask each one when you start, and record it in the basis column. **Record the two legs separately**: once you've judged the four states, add a "which leg cited it" column for each URL. In this trade, the two legs almost never cite the same URL — for the same question, the intersection over union of the two engines' URLs is 17/251 in L3 and 6/106 in family R1. On the same question, the ChatGPT leg might be all government pages while the AI Mode leg is all other firms' explainer articles; combining them into a single intent-level top 10 counts two different battlefields as one. ### Assign an action to each question type ```mermaid Figure: Where law-firm questions go by type; fees, near me and brand go into the first batch, process and rules feed AI Mode only flowchart LR s1["How much does it cost"] --> t1["① ② Fee pages · first batch"]:::hl s2["Which one is near me"] --> t2["⑪ Office page + directory listing"]:::hl s3["Is this firm good · AI got it wrong"] --> t3["Brand question · standalone deliverable"]:::hl s4["Who's best"] --> t4["Off-site: ranking guides and directories"] s5["Process, definition, rules, checklist"] --> t5["⑥⑦⑫⑬⑭ · AI Mode only"] s6["Legal aid · a policy question"] -->|turn into an execution question| t6["⑥ One question, one page"] s7["Changing lawyers · second opinion"] --> t7["⑧ Second-opinion page"] ``` **Fee questions go in the first batch.** This is the door ChatGPT walks through most often to reach law-firm websites. Across L3's 5 price questions, almost everything ChatGPT took from law-firm websites was that firm's own published fees — fixed-fee packages, hourly rates, consultation fees; it folds these into its own market-rate table, and some answers even name the firm. The question on probate legal fees did this both times, in wording such as "S$X nett including court filing/commissioner fees", "S$X before GST and excluding court fees", "S$X all-in": sentences shaped like "price + what's included or not". The AI will quote market rates on its own; a law firm can't control that. What a firm can control is whether its own price shows up, shaped as "service tier + defining conditions + price + what's included or not", in the raw material the AI is pulling its table from. Writing the firm's own fee range [Original text not obtained; see Appendix A.6]: no prohibition found in the current rules (scope checked: the PCR in full [including Part 5 (r 37–49), r 17, r 18 and r 18A], all category 6 PDs and GNs on the Law Society's ethics page, the revoked Publicity Rules (in full); check date: 2026-09-29), 75% confidence, written as in chapter 1, 1.2. Leave peer price ranges out entirely: a range that names a firm or lets it be identified [Statute text; see Appendix A.1]; an unnamed market range [Conservative line (not statute text)] (85% confidence). Not every fee question leaves room for a law firm, either: 2 of L3's 5 price questions cited no law firm in either run — the question on HDB conveyancing legal fees cited HDB, CPF and a bank's legal-fee estimate page; the Chinese-language question on how much a lawyer costs for an employment dispute cited the courts, the Ministry of Manpower or the Legal Aid Bureau, and the Law Society's fee-dispute page. For a fee question with an official or bank-published basis, run it through the four states and count the reachable-enough URLs first, then decide whether to include it. **Near-me questions go in the first batch.** Both L3 run-1 questions cited a law firm, always the contact page or the "find us" page; for the conveyancing near-me question, ChatGPT laid out a table with firm, address, phone and notes columns, with the notes column listing which practice areas each firm handles; one firm's notes were copied from a Law Society directory listing (single case). In run 2, law-firm websites were 0/8, but firm names still made it into the answer, with the source swapped for a Law Society directory listing, a rating directory and an office building's own website; one of these answers even listed a non-law-firm business's contact page as if it were a law firm (single case). So do all three things for near-me questions together: ⑪ the office page (address + nearest MRT station + which practice areas that office handles, on the same line), the Law Society directory listing, and the business profile — keep all three word for word consistent (chapter 5, 5.18; chapter 3, 3.4). The Law Society states that listings in this directory are paid for: fix up any entry you already have first; if you don't have one yet, get the rate card before deciding whether to buy, and do this after the free steps (chapter 6, 6.2). Community legal clinics turn up in this same group too — one L3 near-me question listed the Ministry of Law's published clinic list both times — that's a government page, not reachable, not worth competing for. **Brand questions ("is this firm good", AI getting the firm's own facts wrong) go in the first batch, and take slot 1 this quarter.** For L3's two questions and family's one question, the firm being asked about had its own page cited at least once on each question; across L3's three calls, every law-firm website cited was the firm being asked about (including any keyword microsite it runs). In the answers to these two questions, ChatGPT also read four kinds of third party: the practice areas and lawyer bios on the Law Society directory listing, the department tier in a ranking guide, the rating and review count on a rating directory, and how many published judgments a case-law database shows the lawyer as having handled; once it even warned readers to discount phrases like "charges dropped" and "best possible outcome" on a law firm's own website (L3's criminal question, run 1, single case). So what needs fixing for brand questions is the facts — /facts and the lawyer's own profile page (chapter 3, 3.1–3.2), rankings and directory listings (chapter 3, 3.3–3.4; chapter 6, 6.3) — not another page of self-praise. This is the one item in the ten types where stopping makes things go backwards, and the one cell a law firm can reproduce and see for itself: after the fix, what the AI says often changes first (not guaranteed), and whether it changed shows up in the monthly retest, so make it a standalone deliverable. **The main move for "who's best" is off-site.** Across L3's three questions, ChatGPT run 1 cited only 1 law-firm website (1/12): the criminal "best defence lawyer" question listed individual lawyers by their Doyle's 2026 tier both times, and run 2 additionally cited Legal 500 and Best Lawyers; the personal-injury question cited a "top-rated" list article on a legal-information site and the Law Society directory, with the directory listing's bio paraphrased into the answer. For the family-law question asking for the best divorce lawyer, 4 of 6 citations were Doyle's (×2), Legal 500 and Chambers (single case). Each question got only one or two calls, so this is directional only. The searches ChatGPT ran point the same way: across the 6 answers to these three questions (two runs each), all 6 searched the Law Society directory; only the criminal question searched ranking guides (in both runs), so treat ranking guides first as a move for practice areas that have rankings, such as criminal defence (question-breakdown table at the end of this section). Whether the firm's website is cited and whether the answer names the firm are two different things: across the 10 answers to the 5 "who's best" and "near me" questions (two runs each), 8/10 named specific law firms (each checked by hand), often citing the Law Society directory or a ranking guide rather than the firm's own website; across L3's three runs (ChatGPT twice, AI Mode once), 40 law-firm websites were cited, and 25/40 only once. Three moves: take part in ranking guides through their normal review process, and once listed, the lawyer's own page states only their own entry (chapter 3, 3.3); fill in your directory listing completely (chapter 6, 6.3); for editor-written lists, only supply checkable facts and send correction letters, never pay. **Never build your own ranking list** [Conservative line (not statute text)]: r 43(1)(c) bans any comparison of the quality of service provided by another lawyer [Statute text; see Appendix A.1], and treating a ranking of peers as that kind of comparison is this book's reading of the rule. One law firm's home page in L3, whose title calls itself the best personal-injury firm locally, was still cited by ChatGPT (the personal-injury question, run 2, single case): being cited doesn't mean you may write it; by default, don't call yourself best or No. 1 [Conservative line (not statute text)] (80% confidence). **"Which is better, A or B" only compares procedures, never law firms** — contested vs uncontested divorce, simplified vs normal procedure, probate vs letters of administration, mediation vs going to court; never make any comparison of another firm's fees or quality of service [Statute text; see Appendix A.1]; for how to write it, see chapter 5, 5.7. L3 has only 1 question asking "what's the difference between A and B" with no fee attached (the difference between probate and letters of administration), classed as a definition question, and ChatGPT cited 0 law firms both times; a comparison page on "contested vs uncontested divorce" that does carry fees was cited once by ChatGPT in the family set (single case, page ownership to be verified). Writing the firm's own fee on each side of a comparison page is, so far, the only wording we've seen make it onto the ChatGPT leg. **Process, definition, rules, checklist and calculation questions don't go in the first batch; whatever's left is accepted on the AI Mode leg only.** Across both runs, ChatGPT cited a law-firm website only once in these types combined (checklist, run 2); AI Mode cites law firms anyway — 6 of 7 process questions, 4 of 5 rules questions, 2 of 2 definition questions, 2 of 3 checklist questions, 1 of 1 calculation question. Of the General Edition's ten question types, pure educational "what is it, should I hire one" questions drop out of the pool by default; a law firm's process, definition and rules questions come closest to that cell, but they don't drop out automatically: whether to keep them is judged by check 1 in （→ 通用版 4.5 买家归堆、年机会值与三道关） — look at whether the answer names any firm at all, not whether it cites one. L3 only counted citations, not naming, so this cell should be decided by what you count in your own baseline. Whatever you keep gets written as ⑥⑦⑫⑬⑭ (chapter 5, 5.9, 5.10, 5.12, 5.13); don't expect these to win you position on the ChatGPT leg. When chapter 7 can't fill next month's three tasks, this is the type to thicken first (chapter 7, 7.2). **For policy questions like legal aid, rewrite as an execution question and judge it again.** L3 has 2 legal-aid questions: in ChatGPT run 1, 14 of 15 citations were government and public-body pages, and in AI Mode 8 of 9 were; neither leg cited a law firm. Judging questions like this not reachable straight away would kill off most of a law firm's questions on the ChatGPT leg along with them — 115 of L3 run 1's 146 citations overall were government and public-body pages. Rewrite the question first: turn "do I qualify for legal aid" into "if legal aid isn't approved, how does a lawyer charge for this kind of case, and can I pay in stages", then run it through the four states again; only count it against the abstain line if it's still dominated after the rewrite. The rewritten question goes to ⑥ one question, one page: copy the official eligibility conditions from the Legal Aid Bureau's own text and note the check date (for how to write ⑰, see chapter 5, 5.6); write the firm's own fees as in chapter 1, 1.2, and link across to the ① single-matter fee page; never write a sentence like "come to us if you can't afford a lawyer" (r 39(1) bans touting [Statute text; see Appendix A.1]; applying it to a page like this: [Conservative line (not statute text)]; see chapter 5, 5.6). **Chinese-language questions** go to the same places as English ones, but are accepted on the AI Mode leg only: for L3's 4 Chinese questions, ChatGPT cited a Chinese-language page 0/47 across both runs, AI Mode 5/22; for family's 3 Chinese questions, 1 of 28 citations was a Chinese-language page (AI Mode, single case); Chinese-language pages are judged heuristically, by title and URL. A Chinese-language page is optional — only build one if you have Chinese-speaking clients (chapter 5, 5.3). ### Question breakdown: the pages a law firm needs Use the question breakdown in （→ 通用版 4.7 逐态处置、十类问题与本季名额）. The table below counts, for each kind of question, how many of L3's 60 ChatGPT answers (30 questions × 2 runs, 2026-09-29, family law excluded) searched each category. "Who's best" has only 3 questions, so read it question by question, not as a proportion: | What AI checks | Which questions check it most | Which page or off-site entry answers it | |---|---|---| | The Law Society and directories | Who's best 6/6, is this firm good 4/4, fee questions 6/10 | Fill in the Law Society directory listing completely, word for word the same as the website (6.2, 6.3); office page (5.18) | | Fees | Fee questions 10/10 | Fee guide page (5.4), single-matter fee page (5.5), legal aid and pro bono (5.6) | | Courts and official procedure | Explanation questions such as process and rules 32/36 | One question, one page (5.9), step-by-step procedures and deadlines (5.13), requirements and eligibility page (5.10); this category is mostly held by the courts and ministries | | Reviews | Is this firm good 4/4 | Rating directories and reviews (6.5) | | Ranking guides | Only the criminal "best defence lawyer" question, in both runs | Take part in ranking guides through their normal review process (6.3); the lawyer's own page states only their own entry (3.3) | | Checking each named firm (by firm name or site: website) | Near me 4/4, is this firm good 4/4, who's best 2/6 | The named firm's /facts (3.1), lawyer profile page (3.2), single-matter fee page (5.5), office page (5.18) | This table counts what AI searched for, not what it cited; the citation figures stay on the L3 citation basis used earlier in this section. ### Changing lawyers and second opinions: not in the first batch The General Edition puts second-opinion / remedy questions in the first batch; the Law Firms Edition doesn't follow that — none of L3's 30 questions or family's 22 questions asked this way, so this cell has zero cited samples for law firms. If your firm genuinely handles this kind of work, put one question in the frozen question pool (chapter 2, 2.2) and rank it once your baseline gives you a number. If you do write the page, write ⑧ the second-opinion page; for its skeleton, see chapter 5, 5.11. Writing this page at all [Original text not obtained; see Appendix A.6]: no prohibition found in the current rules (scope checked: PCR 2015 in full, the Law Society Council's 12 Practice Directions and 2 Guidance Notes; check date: 2026-09-29). ```split Figure: On the second-opinion page, the only subject allowed is the client's own situation; write none of the left column, write all of the right column Never write || Write only "The previous lawyer messed it up" || What stage of the process the client is in now, how much time is left "Many clients who come to us find that …" || Whether we can take the case; check for a conflict of interest first Urging the client to end the engagement || Not interfering with the existing engagement Success rate || The scope of the opinion and the fee ``` The four items in the left column don't all rest on the same strength of basis: - Criticising the previous lawyer [Statute text; see Appendix A.1] — r 43(1)(c)'s own text requires "the publicity does not make any comparison or criticism in relation to the fees charged, or the quality of the services provided, by any other legal practitioner or law practice". Never write anything that comments on a previous lawyer's fees or the quality of their work. - An opening like "many clients who come to us find that …" is either talking about another firm's quality of service or about another client's case — write neither [Conservative line (not statute text)] (for how far you may go in describing clients and cases, see chapter 1, 1.4). - Never write a success rate, full stop [Statute text; see Appendix A.1]. - Never urge a client to end their engagement, and don't interfere with an existing engagement [Conservative line (not statute text)] — the basis is the ban on touting in r 39(1): "A legal practitioner or law practice must not tout for business or do anything which is likely to lead to a reasonable inference that the thing was done for the purpose of touting." [Statute text; see Appendix A.1]. Applying it to "urging a client to change lawyers" is this book's reading. No provision specifically governing "urging a client to change lawyers" was found [Original text not obtained]; the earlier draft's line about not improperly influencing a client to end an engagement is no longer cited as a basis. A family law firm's second-opinion page doesn't write up any individual case either. For family law, case studies, republishing reviews and replying to reviews are done case by case, with the written sign-off of a partner or director, but only after passing our own de-identification check first (Appendix B.4): never publish information that could identify a child involved in the proceedings (CYPA s 112), and never publish family-proceedings content a court has restricted from publication (FJA s 10) [Statute text; see Appendix A.4]. These two are statutory provisions: no lawyer's sign-off can release them. Whoever publishes bears the criminal liability personally, and a lawyer's signature cannot shield us from it either. A written sign-off resolves only the stop-and-escalate situations; it never turns something banned as Statute text into something publishable. When talking to a client about this type of question — in fact about the competitive picture for any question type — **you may say only one sentence**: "Who AI names for these questions today, and whose pages it cites as sources, we will count on the spot in the baseline." ❌ Never say "no one is doing this" or "competition is near zero", and never report a specific number of firms: if the client runs their own search next week and finds several, that is a loss of trust. ### This quarter's slots: how the first batch fills the list Once every question has an action assigned, fill this quarter's list following the three combination rules in （→ 通用版 4.7 逐态处置、十类问题与本季名额）: at least one hold-pile question, at least one brand / correction question, and at most one national head term with no location. Here's how a law firm's first batch fills the slots: ```steps Figure: The order a law firm's first batch fills into this quarter's slots; the fee question and near-me question in the same group share one annual opportunity value Slot 1 | Brand question | What AI says about the firm and what it gets wrong; filled as N/A Slot 2 | Hold-pile question | The one with the highest annual opportunity value among questions whose baseline isn't 0 From slot 3 | Take in descending order | By annual opportunity value; within each group, fee and near-me questions go first Last slot | National head term | A "who's best" question with no location, at most 1 slot ``` Hold-pile questions are picked first from fee and near-me questions that have already been cited, or already been named: the baseline isn't 0, so a change can be measured within 90 days. A "who's best" question with no location, like "best divorce lawyer Singapore", counts as a national head term; its main move is off-site, and it takes at most the last slot. Explainer-type process and rules questions come after the fee and near-me questions in the same group. Once the slots are set, the first batch's build order is in chapter 0, 0.1.