A Fable InLead playbook
Get your business mentioned in Claude.
Eight plays you can run this month. Built on our own test of 100 buyer questions. Singapore, Malaysia, the UAE, Saudi Arabia, and the UK and Europe.
Get the PDF.
The whole playbook as a 27-page PDF, free. Plus the tracking template and our 100 questions as CSV files. Share your work email and download it at once.
Your download is ready.
The links work for 7 days. Come back to this page any time to get new ones.
Our study, 17 Oct 2026
Who Claude can find.
100 buyer questions, 8 B2B services, 5 markets. We asked Claude each one three times: twice with web search on, once with no tools. Then we checked what it found, cited and named.
Searched first
91 of 100
questions sent Claude to the web before it answered. All 80 “best” and “recommend” questions did.
Lists and directories
68%
of the 584 links Claude cited in its answers. A firm’s own pages were 30%.
New names
92%
of about 329 firms Claude named after searching never came up in any no-search answer.
Trust warnings
80 of 80
buying answers warned that its sources might be vendor marketing, self-ranked or paid for.
What Claude cited.
All 584 links Claude wrote into its 91 searched answers, by type.
Source: Fable InLead study, 17 Oct 2026 (our analysis). Cross-check run: 72% lists and directories, 27% firm pages.
Our analysis. Claude Sonnet 5.5 via the Claude Code command line tool. Main run and cross-check run, 17 Oct 2026. The cross-check gave the same main findings. Method and limits are in the full playbook.
Eight plays.
Each play has data, an example, exact steps and the tools to use. It also says how long it takes and how to check it worked.
15-minute quick startClaude visibility score30-day planStudy method
Quick start
Fifteen minutes. Four checks.
Do these before anything else. They tell you which of the eight plays to start with.
Ask three buying questions. 5 min
Turn web search on and start a new chat for each. Try “best [service] in [city]”. Try “Can you recommend [service] in [country]?” Then try “Who can [result] for a [buyer]?” Are you named?
Sort the sources it shows. 4 min
Open the source links under each answer. Mark each one: a list, a directory, or a firm’s own page. This is your target list.
Check your robots.txt. 2 min
Open yourdomain.com/robots.txt. Make sure no “Disallow” line blocks ClaudeBot, Claude-User, Claude-SearchBot, Googlebot or all bots (*).
Look yourself up. 4 min
Search your company on Clutch and on two directories from step 2. Missing or half-empty profiles are your first job this week.
| What your check showed | Start with |
|---|---|
| Claude never searched for your questions | Play 1, then play 7 |
| robots.txt or your host blocks a Claude bot or Googlebot | Play 2, today |
| The sources are mostly directories, and you are not on them | Play 3 |
| The sources are mostly “best” lists, and you are not on them | Play 4 |
| Firms’ own city pages show up, and you have none | Play 5 |
| You are named, with no proof or a wrong fact | Plays 6 and 8 |
Updated 17 Oct 2026
First edition.
Study run on 17 Oct 2026 with Claude Sonnet 5.5. Claude changes often, so every number here carries its date. We plan to re-run the same 100 questions on 17 Nov 2026.
Sort your buyer questions by what Claude does with them.
Claude looks some questions up and answers others from memory. You can win the first kind with work you do this month. The second kind takes longer.
Why it matters
When Claude searches, it reads pages that exist today. A new profile or page can start to count soon after it is indexed. When Claude answers from memory, it uses what it learned in training. That changes only when the model is updated.
In our test, all 80 “best” and “recommend” questions sent Claude to the web. None of the 5 “how do I choose” questions did. 3 of 5 “agency or in-house” questions did not. Searchable’s July 2026 study also found that wording matters. Claude searched “best CRM 2026” 9 times in 10. It searched “how to pick a CRM” 1 time in 34.
How often Claude searched, by question type.
Questions where Claude ran a web search before answering. 91 of 100 overall.
Source: Fable InLead study, 17 Oct 2026, 100 questions (our analysis).
What it means for a B2B service firm
You can move fastest on the “best” and “recommend” questions. Plays 2 to 6 cover them. The “how do I choose” questions take longer to win. Play 7 covers those.
Time
45 minutes
Tools
Claude with web search on. A spreadsheet, or our tracking template (play 8).
Check it worked
You have 20 questions in two columns, with the names and sources Claude gave for each.
Do this
- Write 20 questions your buyers ask, in their words. No brand names. Mix them: about 8 “best”, 6 “recommend”, 3 on cost and 3 “how do I choose”.
- Add Arabic or Malay versions if your buyers use them. Gulf buyers may ask in Arabic. Our study used English only, so test both.
- Open Claude and check web search is turned on. Start a new chat for every question, so one answer does not shape the next.
- Ask each question and look under the answer. Source links mean Claude searched. No sources means it answered from memory.
- Sort them into two columns: searched, and memory. Note who Claude named and which sites it linked. Searched questions go to plays 2 to 6. Memory questions go to play 7.
Example: questions from our study, as a firm in Kuala Lumpur or Singapore could track them
| Question, in the buyer’s words | Type | Searched in our test? |
|---|---|---|
| best B2B lead generation agency in Malaysia | best | Yes |
| Can you recommend good LinkedIn outreach agencies in Malaysia for a B2B tech company? | recommend | Yes |
| How much does a B2B lead generation agency cost in Malaysia? | cost | Yes |
| Lead generation agency vs in-house SDR team in Malaysia: which is better for a B2B company? | vs | Yes |
| How do I choose a B2B lead generation agency in Malaysia? | choose | No |
| Lead generation agency vs in-house SDR team in Singapore: which is better for a B2B company? | vs | No |
Open the door to Claude’s crawlers.
Claude’s search can only use pages it can reach. Three Anthropic bots and Brave’s crawler need a clear path to your site.
Why it matters
Anthropic names three bots in its help centre (page updated 7 April 2026). ClaudeBot collects pages for training. Claude-User fetches pages when someone asks Claude a question. Claude-SearchBot indexes pages for Claude’s search. Anthropic says blocking the last two may cut your visibility in Claude’s answers.
Claude’s search has been linked to Brave Search since March 2025. That month Anthropic added Brave to its list of subprocessors, as TechCrunch and Simon Willison reported. In July 2026, Xponent21 reported Brave still on that list. It also reported a second web search subprocessor, turbopuffer, whose exact role is not public.
Brave’s crawler does not announce itself. Brave says its bot skips any page Googlebot cannot crawl. One wrong line in robots.txt can hide you from Claude. So can one “block AI bots” switch at your host. Google may still show you, so you may never notice.
Time
20 minutes
Tools
A browser, Brave Search, curl or View Source, your host’s dashboard.
Check it worked
No Disallow for those bots. Your text shows in the raw HTML. site: on Brave lists your pages.
Do this
- Open
yourdomain.com/robots.txt. Look forDisallow: /underUser-agent: *,Googlebot,ClaudeBot,Claude-UserorClaude-SearchBot. - Add the four allow blocks from the robots.txt example if you want to be found. Keep any rules you need for private folders.
- Check your host or CDN bot settings. Many hosts have a one-click switch that blocks AI crawlers. Make sure it lets the bots above and Googlebot in.
- Check your key pages work without JavaScript. Run
curl -s https://yourdomain.com/your-page | grep -i "your service". No match means JavaScript builds the text. Not every AI crawler runs it. - Check Brave can see you. Search
site:yourdomain.comon search.brave.com. Nothing there? Submit your homepage atsearch.brave.com/submit-url.
Example: excerpt from fableinlead.com/robots.txt (live, 17 Oct 2026)
User-agent: Googlebot Allow: / User-agent: ClaudeBot Allow: / User-agent: Claude-SearchBot Allow: / User-agent: Claude-User Allow: /
Get on the directories Claude’s search keeps finding.
Directories were the second biggest source in what Claude’s search returned. Which ones matter depends on your market.
Why it matters
Directory links were 30% of what Claude’s search returned. They showed up in 70 of the 91 questions it searched. Claude cited 45% of the directory links it got back.
One directory came up by name more than any other: Clutch. It was named in 53 of 100 answers with search on. With no search at all, it was named in 51.
The other directories differ by market. In Saudi Arabia, Semrush Agency Partners pages came up most. In Singapore and the UAE, it was the agency list of Saleshandy, a sales software company. In Malaysia it was Clutch.
What Claude said when profiles were thin (its own words, 17 Oct 2026)
“The results were almost all vendors’ own pages, and I found no independent reviews, such as Clutch profiles.” (best LinkedIn outreach agency in Saudi Arabia)
“The results came mostly from directory sites, which often include paid placements, and my Clutch search returned nothing for Saudi Arabia.” (recommend B2B marketing agencies in Saudi Arabia)
Which directories Claude’s search returned, by market.
Searched questions where each directory showed up at least once (about 18 per market).
| Market | Showed up most (searched questions) |
|---|---|
| Singapore | Saleshandy agency list 9 of 18 Semrush Agency Partners 3 of 18 |
| Malaysia | Clutch 9 of 19 Salesforge agency directory 5 of 19 GoodFirms 5 of 19 |
| UAE | Saleshandy agency list 11 of 18 Salesforge agency directory 8 of 18 Semrush Agency Partners 6 of 18 |
| Saudi Arabia | Semrush Agency Partners 12 of 18 TechBehemoths 7 of 18 Clutch 5 of 18 |
| UK and Europe | No clear leader. Semrush and Saleshandy 2 of 18 each. Seven others, including Sortlist and Clutch, 1 each. Start with play 4. |
Source: Fable InLead study, 17 Oct 2026, links returned by Claude’s web search (our analysis). UK and Europe questions returned few directories; lists led there (play 4).
Time
2 to 3 hours, then reviews over a few weeks
Tools
Clutch first, then the directories your own Claude runs showed.
Check it worked
Re-ask your play 1 questions next month (play 8). Look for your profile link in Claude’s sources.
Do this
- List every directory link Claude showed you in play 1. Add the top ones for your market from the table above.
- Start with the free profiles. Clutch first. Weigh the paid programs against what they cost you. Some directories charge for a profile. That is different from paying a list author for a top spot.
- Fill every field the same way. Use the same name, one-liner, services, cities, founding year and price range. Word for word, on every site.
- Pick the right city and service category. Claude’s search returned pages like Semrush’s list of lead generation agencies in Riyadh. Be listed under the city your buyer types.
- Ask 3 to 5 clients for a Clutch review. Reviews came up in 65 of 80 buying answers.
How to get on each one (checked 17 Oct 2026)
| Directory | How you get listed |
|---|---|
| Clutch | Free basic profile (clutch.co, “Get listed”). Verified client reviews make it count. |
| GoodFirms, TechBehemoths, DesignRush | Each offers a company profile form on its site. Free tiers are reported by third parties; confirm on sign-up. |
| Semrush Agency Partners | A paid Semrush plan, the Agency Growth Kit and two Semrush exams (its terms of use). |
| Saleshandy agency list | Needs a paid Saleshandy plan (from US$99 a month) to join its Agency Partner Program. |
| Salesforge agency directory | Its list sits at salesforge.ai/directory/agencies. No public sign-up page found. Ask through its demo or contact form. |
Part two
The full playbook.
Plays 4 to 8 and the 20-minute visibility score. Plus a 30‑day plan, the tracking template, our 100 questions and the method.
Earn a place on the “best agencies in [city]” lists.
Lists were 41% of the links Claude cited. Claude warns about them. It still builds its shortlist from them.
Why it matters
“Best agencies” lists were the biggest single source. They were 36% of the links Claude’s search returned, and 41% of the links it cited. Once Claude got a list back, it cited it 59% of the time. Firm pages came close at 55%. Directories were lower at 45%.
Claude also reads lists with care. Every one of the 80 buying answers gave a warning. Its sources might be vendor marketing, self-ranked or paid for. In 20 answers it said a list ranked its own publisher. It still built the shortlist from them. A second run with a plain system prompt showed the same warnings in 80 of 80.
One host supplied the most. Salesforge, a sales software company, supplied 127 of the 406 list links, from 35 pages. 67 of those links were translated copies. Check its city lists for an author or contact link. If there is none, move on to the next list.
41%
of the links Claude cited in its answers were “best” lists.
59%
of the list links Claude’s search returned were cited in the answer.
236
different list pages, from 123 sites. 125 list links had 2026 in the title or link.
Time
3 hours to start, then 30 minutes a week
Tools
Brave Search, Google, a spreadsheet, email.
Check it worked
Each month, check the lists Claude cites for your city (play 8). Aim to be named on one within two months. Our estimate, not a measured figure.
Do this
- Collect the list pages for your service and city. Start with the sources from play 1. Then search Brave and Google for
best [service] in [city] 2026. - Score each list. Updated in the last 6 months? Names firms of your size? Has a contact or “suggest a company” link? Keep the ones with three yeses.
- Send the author the example note. One fact they can check, one review link. Follow up once after a week.
- Do not pay for a spot. Claude flagged sponsored or paid placements in 14 answers.
- If you publish your own “best” list, do not rank yourself first. Claude pointed this out in its answers. Write an honest comparison with real facts on each firm.
Claude’s answers, 17 Oct 2026, in its own words
“It puts itself first on its own list.” (best B2B marketing agency in Singapore)
“Treat the awards as a better signal than the directory rankings, which can be influenced by sponsored placements.” (best B2B marketing agency in Malaysia)
A note to send a list author (under 50 words)
Hi [first name], I read your list of [service] firms in [city]. We are a [service] firm in [city]. [One checked result with a number.] Our Clutch reviews: [link] Happy to send anything you need to check us. [Your name]
Build one page per service, per market.
Firm pages that name the city or country in the link got cited more. One page per service and city gives Claude a clear match.
Why it matters
When Claude’s search returned a firm’s own page, it cited it 55% of the time. Pages with the city or country in the link were cited 61% of the time. Without it, 46%.
58% of the firm-page links had the city or country in them, like /b2b-lead-generation/riyadh. One site with city pages for every service supplied 72 of those. Without that site, the cite rates were 66% and 47%. It is a correlation. It does not prove cause.
Local web addresses were rare. Of 281 links returned for Saudi questions, 0 were on a .sa domain. For Singapore it was 17 of 189 on .sg.
The Ahrefs AEO course gives three writing rules for pages like this. Answer first. Make each section stand alone. Name real places, buyers and services.
Firm pages with the place in the link were cited more often.
Share of returned firm-page links Claude cited. 163 had the place in the link, 119 did not. Without the biggest city-page site: 66% and 47%.
Source: Fable InLead study, 17 Oct 2026 (our analysis). Correlation only.
Time
1 day per page
Tools
Your site, Google Search Console, Bing Webmaster Tools, Brave Search.
Check it worked
The page shows for site:yourdomain.com [city] on Brave. Then watch for it in Claude’s sources each month (play 8).
Do this
- List your services and the markets you sell to. Each pair a buyer would search for gets one page. Start with the two pairs that bring the most revenue.
- Use the link and title pattern in the example below. Service first, then the place, in plain words buyers use.
- Write the first two lines as the answer. Claude may quote only those two lines. They should still say what you do, where and for whom.
- Add one real result, one review link and a price range. Play 6 shows why.
- Link the page from your services page and add it to your sitemap. Then request indexing in Google Search Console and Bing Webmaster Tools. Check it with
site:on Brave.
Example: outline of one service and market page
| Part | What goes in it |
|---|---|
| Link | /linkedin-outreach/dubai |
| Title | LinkedIn outreach for B2B firms in Dubai |
| First two lines | What you do, for whom, where, and one checked result. |
| Who it is for | Buyer roles and industries in that market, by name. |
| How it works | Four or five plain steps. |
| Proof | One case with numbers, one client review link. |
| Price | A real range, or what it depends on. |
| Local rules | The data law that applies (UAE PDPL, for example), in one paragraph. |
| Updated | A visible date that changes when the page changes. |
Put your proof where Claude can quote it.
Claude kept mentioning reviews, case studies and prices. Firms that publish them give Claude something to quote.
Why it matters
Claude kept telling buyers what it could not find. In 42 of 80 buying answers it said it found no independent ranking or review. Reviews came up in 65 of them, pricing in 55 and case studies in 52.
Proof on a public page gives Claude something to quote. Proof that only lives in a sales call gives it nothing.
Time
Half a day to start
Tools
Your site, Clutch, G2 or Google Business Profile, your own client results.
Check it worked
Re-run play 1. Claude’s answer about you mentions a result, a review or your price range.
Do this
- Publish two or three case studies with numbers. Name the client if they agree, or describe them by type. Give the market, what you did and the result in meetings, leads or revenue.
- Get reviews where Claude looks. Clutch first. Then G2 or your Google Business Profile if your buyers use them.
- Show a price range. A cost page in the buyer’s currency answers the cost question. Claude searched all 5 cost questions in our test.
- Add one page on data law. Say how you handle personal data in outreach. Singapore and Malaysia: PDPA. UAE and Saudi Arabia: PDPL. UK and EU: GDPR. Get legal advice for your own case.
- List awards and named partners, with a link to the source. Awards came up in 5 answers. In one, Claude told the buyer to treat awards as a better signal than directory rankings.
What Claude mentioned in its buying answers.
Number of the 80 “best” and “recommend” answers that use each word or phrase.
Source: Fable InLead study, 17 Oct 2026. Keyword count in Claude’s answers with web search on (our analysis). A mention is not always a request.
Win the questions Claude answers from memory.
Without search, Claude’s most-named firms were big international names. With search, most of the firms it named were new. Memory changes slowly.
Why it matters
With no search, Claude named at least one firm in 64 of 100 answers. In 17 answers it said it would not name firms from memory. Its most-named firms were well-known international outbound agencies, global ad agency networks and call-centre groups.
With search on, it named about 329 different firms, and about 220 of them only once. 302 of them never came up in any no-search answer. For a boutique firm that is good news: search gives you a way in.
News pages were rare in Claude’s search results here: 6 of 1,138 links. So in these B2B service searches, press helps mostly over time, through memory. Searchable’s larger study found Claude cited national press about half the time it read it. The Ahrefs AEO course names one strong signal from a study of 75,000 brands. Brands mentioned more across the web showed up more in Google’s AI Overviews. That is a correlation, measured on Google. It does not prove cause, and it is not a Claude figure.
Search brings in new names.
Same 100 questions, asked with web search on and with no tools.
Answers naming at least one firm (of 100)
Different firms named (approximate, after merging spellings)
Source: Fable InLead study, 17 Oct 2026. Firm names read from each answer by a second model, then merged for spelling (our analysis).
Time
1 to 2 hours a week, for months
Tools
Your site, LinkedIn, the media list below, podcasts and event speaker pages.
Check it worked
Ask Claude with web search off: “What is [your firm]?” It describes you correctly. Later, it names you in a memory question.
Do this
- Write one clear guide for each memory question from play 1. For example, “How to choose a B2B lead generation agency in Malaysia”. Answer in the first lines, then the detail.
- Give your method a name and use it everywhere. Same name on your site, LinkedIn, talks and podcasts.
- Keep one set of facts about your firm. Same name, one-liner, founder, founding year and services on your site, LinkedIn, directories and bios.
- Pitch one outlet a week. A guest article, a data point from your own work, or an expert comment. Log each pitch in your tracking sheet.
- Test memory every quarter. Ask Claude with web search off: “What is [your firm]?” and your memory questions. Log what it says.
Where to get mentioned: examples by market
| Market | Examples of business media to pitch |
|---|---|
| Singapore | The Business Times, e27, Tech in Asia |
| Malaysia | The Edge Malaysia, The Star (Business), Vulcan Post |
| UAE | Gulf News, Khaleej Times, Arabian Business |
| Saudi Arabia | Arab News, Argaam, Saudi Gazette |
| UK and Europe | City AM, Sifted, Marketing Week |
A pitch to an editor (under 50 words)
Hi [first name], I read your piece on [topic] in [outlet]. We run [service] for B2B firms in [city]. One number from our own work: [checked figure]. Happy to send a short comment or the data behind it. [Your name]
Track it every month.
Ask the same questions every month. Count how often you are named, and where Claude found you.
Why it matters
Answers change between runs. The Ahrefs AEO course says to count how often you are named across many runs. So keep the same questions, ask each one three times, and compare month to month.
Test Claude itself. A search engine rank is only a rough guide. Brave let us collect results for 13 Singapore questions. Only 13% of the links Claude’s search returned were in Brave’s top 10 there. Claude often rewrote the question before it searched. The sample came from one location with no country setting. Searchable’s much larger July 2026 study found a 63% overlap with Brave’s top 10. It measured the pages Claude reads, a different measure. Use Brave to check you are indexed, and Claude to check you are named.
Time
1 hour a month
Tools
Claude, the tracking template (CSV), your web analytics, a booking form question.
Check it worked
A dated log of the same questions, month after month. A trend you can show your team.
Do this
- Copy the tracking template. Log one row per question per run. See the template below the 30-day plan.
- Run your 20 questions three times each, on the same day each month. New chat every time, web search on. Your mention count is how many of the 60 answers name you.
- Log every source domain. New domains are your next directory or list targets (plays 3 and 4).
- Ask “What is [your firm]?” every month. Fix any wrong fact on your own site first, then on the profile Claude quoted.
- Watch for visits from claude.ai. In Google Analytics 4: Reports, Acquisition, Traffic acquisition. Filter session source for “claude”. Many AI mentions carry no link. So also add “AI assistant” to your “How did you hear about us?” question.
Self-check
Your Claude visibility score.
Twenty minutes, 100 points. Give yourself the points only when the check passes today. Re-score every month.
| # | Check | Points |
|---|---|---|
| 01 | Claude names you in one of 3 “best” questions for your service and city. | 15 pts |
| 02 | Claude links your site in any of those 3 answers. | 5 pts |
| 03 | Claude describes your firm correctly when asked “What is [your firm]?” | 10 pts |
| 04 | robots.txt and your host let ClaudeBot, Claude-User, Claude-SearchBot and Googlebot in. | 10 pts |
| 05 | Your main service page text shows in the raw HTML (curl or View Source). | 5 pts |
| 06 | site:yourdomain.com on Brave shows your pages. | 5 pts |
| 07 | Full profiles on 3 or more directories Claude showed (5 points for 1 or 2). | 10 pts |
| 08 | 3 or more client reviews on Clutch or another directory Claude showed. | 10 pts |
| 09 | At least one “best” list that Claude cited for your service and city names you. | 10 pts |
| 10 | A page for your main service in your main city, with the city in the link. | 5 pts |
| 11 | A price range or a cost page. | 5 pts |
| 12 | Your name, one-liner and founder match on your site, LinkedIn and directories. | 5 pts |
| 13 | Your key pages show a real update in the last 90 days. | 5 pts |
| Total | 100 pts |
0 to 30
Hidden.
Start with plays 2 and 3 this week.
31 to 55
Findable.
Claude can reach you. Work on plays 3, 4 and 6.
56 to 80
In the running.
Push lists and proof (plays 4 to 6).
81 to 100
Strong.
Hold it with play 8 and grow memory (play 7).
Our own scoring guide, built from the study and the Ahrefs AEO course. The weights are our judgement.
Want help with your score? Book 30 minutes with Saif.
30-day plan
One month, four weeks.
About 25 to 30 hours over the month, by our estimate. One owner, one shared sheet, the same 20 questions at the start and the end.
Week 1
Baseline and doors.
- Write your 20 questions (play 1).
- Run each 3 times and log it in the template.
- Fix robots.txt, host bot settings and JavaScript-only text (play 2).
- Check
site:on Brave; submit if missing. - Take your visibility score.
Week 2
Directories and reviews.
- List the directories Claude showed for your market (play 3).
- Claim and complete 3 to 5 profiles with the same facts.
- Ask 3 to 5 clients for a Clutch review.
- Make your name and one-liner match everywhere (play 7).
Week 3
Pages and proof.
- Publish one service and market page (play 5).
- Publish or update one case study with numbers.
- Add a price range or cost page (play 6).
- Request indexing in Google Search Console and Bing Webmaster Tools.
Week 4
Lists and a re-check.
- Score the list pages for your city and pitch the best 5 (play 4).
- Write one “how to choose” guide for a memory question (play 7).
- Re-run all 20 questions 3 times.
- Re-take the score and compare with week 1.
Template
The tracking template.
One row per question per run, so 60 rows a month for 20 questions. The example row shows the format only.
Download the template (CSV)Download our 100 study questions (CSV)
All 100 study questions are in the appendix. Ask us for the template and the questions as CSV files.
| Column | What to write | Example row |
|---|---|---|
| Run date | The day you ran it | 17 Nov 2026 |
| Question ID | Fixed code, never reused | Q01 |
| Exact words | Copied, never edited between months | best B2B lead generation agency in Malaysia |
| Market | Where the buyer is | Malaysia |
| Service | What they buy | Lead generation |
| Type | best, recommend, cost, choose, vs, who | best |
Show the other 10 columns
| Column | What to write | Example row |
|---|---|---|
| Run no. | 1, 2 or 3 | 1 |
| Claude searched? | Source links shown = yes | Yes |
| You named? | Yes or no | No |
| You linked? | Yes or no | No |
| Your position | 1st, 2nd ... name in the answer | Not named |
| Competitors named | As written in the answer | (copy the names) |
| Source domains | Every domain under the answer | (copy every domain) |
| What Claude said about you | Quote it | - |
| Correct? | Yes, no, or what is wrong | - |
| Next action | One action, one owner | Complete Clutch profile (play 3) |
Common mistakes
Nine ways to waste the month.
- Checking Google only. Claude’s search has been tied to Brave. In our Singapore check, only 13% of the links it returned were in Brave’s top 10. Test Claude itself.
- Treating one answer as a rank. Answers change between runs. Ask each question three times and track the share of answers that name you.
- Blocking Claude by accident. A
Disallowunder*or Googlebot can hide every page. So can a host switch that blocks AI crawlers. - Hiding text behind JavaScript, tabs or PDFs. If curl cannot see your service text, some crawlers cannot either.
- Paying for list spots. Claude flagged sponsored or paid placements in 14 answers. Earn the place with a fact the author can check.
- Ranking yourself first on your own “best” list. Claude pointed out self-ranking in 20 answers and told buyers to vet those lists.
- Different facts on different sites. Two founding years or two client counts can confuse AI answers. The Ahrefs AEO course makes the same point.
- Spending your first month on Reddit. For these B2B service questions, Claude’s search returned 1 forum link in 1,138. Start with directories, lists and your own pages.
- Expecting llms.txt to do the work. Ahrefs found 97% of llms.txt files got no requests in a month. No major AI company officially supports it. Keep one if you like.
A note from the founder
I have led 200+ B2B outbound projects. This year we wanted one answer. When a buyer asks Claude for a firm like ours, what does Claude read first?
Most guides we found were built on software brands in the US. This study covers the markets we work in. Those are Singapore, Malaysia, the UAE, Saudi Arabia, and the UK and Europe. We kept every raw file. Ask us and we will share them.
Our own site was not named in any answer. We started where most firms start. We plan to run these eight plays on fableinlead.com too. We will share our own score in the next edition.
Saiful Islam SaifFounder, Fable InLead (fableinlead.com)
Study method and limits
How we ran it.
One day, 100 fixed questions, three Claude runs and a Brave check. Every script and raw file is kept, so the study can be run again. Ask us for them.
The questions
5 markets: Singapore, Malaysia, the UAE, Saudi Arabia, and the UK and Europe. 8 B2B services: lead generation, LinkedIn outreach, appointment setting and B2B marketing. The other four were SEO and AI search (GEO), cold email, outsourced SDR and B2B website design.
Each market got 20 questions. 8 were “best [service] in [market]”. 8 were “Can you recommend ...”. 4 were research questions: cost, how to choose, agency vs in-house, and who can book meetings. Wording was fixed before any run. All questions were in English.
Check 1: Claude with web search
Each question went to Claude Sonnet 5.5 on 17 Oct 2026. We used the Claude Code command line tool, version 2.1.302. Only its web search tool was on, and that tool also used Claude Haiku 5.5. No other tools, settings or plugins ran.
We saved the 147 searches Claude wrote and every final answer. We also saved all 1,340 links the searches returned. After removing repeats, 1,138 links were left, pointing to 786 different pages. The 584 cited links are the links Claude wrote into its 91 answers.
Cross-check: a plain system prompt
The Claude Code tool adds its own instructions. So we ran all 100 questions again with a plain system prompt. The search tool stayed the same. Claude searched 87 of 100. That included all 80 buying questions and no “how do I choose” question. It searched fewer of the other research questions than in the main run. 80 of 80 buying answers carried a warning. With no search at all, only 14 of 80 did. Lists and directories were 72% of the 481 cited links, and firm pages 27%.
A second Claude model read each buying answer for a warning, in both runs. It agreed with our hand read on all 80 answers of the first run.
Check 2: Claude with no tools
The same 100 questions, with every tool off, so Claude could answer only from memory.
Check 3: Brave Search
We collected Brave’s web results one question at a time. Brave limited automated requests, so we kept the 13 questions with clean results. All were Singapore questions. We used them only to compare with Claude’s links.
Reading the results
Rules on each web address sorted every link into one type. Types were lists, directories, firm pages, news, reviews, forums and other. A hand check of 60 random links found 59 sorted right. We fixed a few rules after it.
A second Claude model (Haiku) read each answer and listed the firms and directories it named. We merged spelling variants and spot-checked 4 answers by hand. We looked for warnings about vendor content by keyword, then read every buying answer in full. 80 of 80 carried one.
The 92% new-names figure counts different firms. 302 of about 329 never came up in any no-search answer. Compared answer by answer, 92.4% of names were new where both runs named firms. Across all questions it was 94.1%.
Limits
One main run and one cross-check run per question, on one day. Claude’s answers change from run to run. We tested Claude through Claude Code, with its own and a plain system prompt. The claude.ai app may search and cite differently. The Brave check covers 13 Singapore questions only. With about 18 searched questions per market, treat market rows as signals. For UK and Europe, “best” and research questions said “the UK”, and “recommend” questions said “Europe”. We plan to add a Gemini check in the next edition. 100 questions show patterns for these markets and services. They cannot rank firms, and this playbook names no agency.
Sources
Where every number comes from.
- Fable InLead, “Who Claude can find: 100 B2B buyer questions in 5 markets”, run 17 Oct 2026. Raw data, scripts and method kept with the study.
- Anthropic Help Center, “Does Anthropic crawl data from the web, and how can site owners block the crawler?”, updated 7 Apr 2026. support.claude.com/en/articles/8896518
- TechCrunch, “Anthropic appears to be using Brave to power web search for its Claude chatbot”, 21 Mar 2025. techcrunch.com/2025/03/21/anthropic-appears-to-be-using-brave-to-power-web-searches-for-its-claude-chatbot/
- Simon Willison, “Anthropic Trust Center: Brave Search added as a subprocessor”, 21 Mar 2025 (change dated 19 Mar 2025). simonwillison.net/2025/Mar/21/anthropic-use-brave/
- Isaac Marcuson, Xponent21, “Anthropic Lists Two Web-Search Subprocessors Behind Claude ...”, 2 Jul 2026, updated 16 Jul 2026. xponent21.com/insights/claude-web-search-brave-turbopuffer/
- Brave Search Help, “Brave Search crawler”, read 17 Oct 2026. search.brave.com/help/brave-search-crawler
- Brave Search, Submit URL page, live 17 Oct 2026. search.brave.com/submit-url
- Searchable, “7 Ways To Get Your Business Mentioned In Claude”, study dated July 2026 (plays 1 and 8). www.searchable.com/claude-visibility-report
- Ahrefs Academy, Answer Engine Optimization course by Sam Oh, lessons 1.1, 3.1, 3.2, 3.4 and 4.2, published Apr to May 2026, read 17 Oct 2026. ahrefs.com/academy/aeo-course
- Ahrefs, AI Overview brand visibility factors: 75,000 brands, 0.664 correlation for branded web mentions (course lesson 3.2). ahrefs.com/blog/ai-overview-brand-correlation/
- Ahrefs, llms.txt study of 137,210 sites (course lesson 3.4). ahrefs.com/blog/llmstxt-study/
- Semrush Agency Partners terms of use (certification steps), read 17 Oct 2026. agencies.semrush.com/terms-of-use/
- Saleshandy, Agency Partners program page, read 17 Oct 2026. www.saleshandy.com/become-an-agency-partner
- Clutch, getting listed (free profile), read 17 Oct 2026. clutch.co
Want us to run it for you?
Fable InLead (fableinlead.com) runs AI search visibility work for B2B service firms. We run it alongside our outbound work. We build it around your market, your buyers and your proof.
saif@fableinlead.com · fableinlead.com
Honest by design. Boutique by choice.
Keep a copy.
The same playbook as a 27-page PDF, to read offline or share with your team. Plus the tracking template and our 100 questions as CSV files.
Your download is ready.
The links work for 7 days. Come back to this page any time to get new ones.
More playbooks.
A Fable InLead playbookThe 2027 B2B GTM playbookWhere pipeline comes from next year, and 30 plays to run now. For B2B founders and sales leaders.Read the playbook
A Fable InLead playbookThe complete AEO playbookGet your brand into AI answers on ChatGPT, Google AI Overviews and AI Mode, Gemini, Perplexity, Copilot and Claude.Read the playbook