A B2B site added 34 pages, 12 of them comparison listicles. ChatGPT referrals went from 2 to 46 per fortnight

Findings

1. ChatGPT referral visitors rose from 2 in the 14 days from 17 to 30 August 2026 to 46 in the 14 days from 7 to 20 September, in windows matched by weekday, while visitors from all referral sources rose 3.6% (795 to 824).
2. 40 of the 62 ChatGPT visitors from 3 to 22 September (64.5%) entered the site on one of the 34 new pages: 20 on comparison listicles and 14 on service pages.
3. A daily tracker found no citation of the site in 1,506 AI answers from 27 July to 5 September. From 6 to 9 September, ChatGPT cited it in 4 of 40 answers, each time linking a new comparison listicle.
4. 33 of the 34 new pages were indexed when inspected on 22 September, and 32 of them drew their first Google impression on the day they went live or the next day.
5. Long conversational queries made up 21.0% of the new pages' visible Google impressions from 3 to 22 September (669 of 3,181), at an average position of 6.9 against 51.0 for keyword queries.
6. Google clicks rose 41.0% between the same two windows (134 to 189), but older pages edited on the same day supplied 38 of the 54 extra page-level clicks, so the causes cannot be separated.
7. Three measures showed no gain: two generic "best agencies" prompts named the company in 0 of 10 answers on 22 September, weekly clicks on brand-name searches stayed inside their June to August range, and the new pages drew 26 Bing impressions through the week of 18 September.
8. Asked through its API on 22 September, ChatGPT returned 7 of 13 answers without a single citation, and both of its answers on who founded the company named a founder the company never had.

What the site changed

The site belongs to a US business-to-business services company, a done-for-you lead-generation agency founded in 2022. On 3 September 2026 it published 32 new pages. It published a 33rd on 5 September and a company facts page on 22 September, for 34 in all.

Every listicle said near the top that the publisher is one of the companies on the list, and each described how the vendors were compared and on what date their websites were read. In the 12 listicles as published, the publisher's own position ranged from 3rd to last, with last place on 4 lists and a median position of 10, on lists of 9 to 16 vendors (listicle-self-ranking.csv).

The site changed older pages on the same day. It set 164 near-duplicate template pages to noindex, merged 8 duplicate pages into stronger ones, rewrote the titles of 15 older product pages that already ranked, removed typed-in counts and coverage claims its data did not support from the template behind those product pages, and redirected alternate forms of every URL to one address.

ChatGPT referrals rose from 2 to 46 per fortnight

Chart data

week_start	ChatGPT referral visitors
2026-06-01	1
2026-06-08	0
2026-06-15	0
2026-06-22	2
2026-06-29	3
2026-07-06	1
2026-07-13	2
2026-07-20	2
2026-07-27	1
2026-08-03	1
2026-08-10	2
2026-08-17	2
2026-08-24	0
2026-08-31	9
2026-09-07	27
2026-09-14	19

Source: Strandline Research analysis of the site's Plausible analytics. Weeks run Monday to Sunday and are labeled by their Monday. The new pages went live in the week of 31 August. Weekly values are sums of daily visitors.

In the 13 weeks from 1 June to 30 August 2026, ChatGPT sent the site between 0 and 3 visitors a week. In the week the pages went live, 31 August to 6 September, it sent 9. In the two weeks after, it sent 27 and 19.

The main comparison uses two 14-day windows that each run Monday to Sunday: 17 to 30 August before the launch and 7 to 20 September after it. ChatGPT referral visitors went from 2 to 46. Visitors from all referral sources rose 3.6% over the same windows (795 to 824), so ChatGPT's share of visitors went from 0.3% (2 of 795) to 5.6% (46 of 824). A 7-day check with the same weekday matching, 24 to 30 August against 14 to 20 September, gives 0 against 19. Visitors referred by other AI assistants (Claude, Gemini, Copilot, Perplexity and DeepSeek together) went from 1 to 6 in the same two windows.

Over a longer span, the site had 18 ChatGPT visitors in the 94 days from 1 June to 2 September and 62 in the 20 days from 3 to 22 September. Of those 62, 3 reached the page shown after a contact form is sent, against 1 of the 18 before. Those counts are too small to compare as rates.

Most ChatGPT visitors entered on the new pages

40 of the 62 ChatGPT visitors from 3 to 22 September (64.5%) landed first on a new page: 20 on comparison listicles, 14 on service pages, 3 on guides and research and 3 on industry pages. The other 22 landed on older pages, mostly research pages (10) and product pages (7). The homepage took 7 of the 18 ChatGPT visitors from 1 June to 2 September and 3 of the 62 from 3 to 22 September.

The tracker's first citations of the site went to the new listicles

A daily tracker asked 10 buyer prompts that do not name the company (prompts.csv) to ChatGPT, Perplexity and Gemini from 27 July 2026, and to Claude from 29 July, once per engine per day, until its schedule was switched off after 9 September. In 39 runs from 27 July to 5 September it collected 1,506 answers. Two of them named the company. None cited its website.

The first citations came three days after the launch. From 6 to 9 September, ChatGPT named the company in 5 of its 40 answers and cited the site in 4, each time linking one of the new comparison listicles. Perplexity, Gemini and Claude neither named nor cited it in those 4 runs (100 answers). When we re-ran the same 10 prompts on 22 September, ChatGPT named the company and cited a new listicle in 2 of 10 answers, Claude cited a new listicle in 1 of 10 without naming the company, and Perplexity and Gemini did neither (tracker-daily.csv, ai-scan.csv).

Google indexed 33 of the 34 pages within a day

Comparison listicles4,615

Industry pages625

Guides and research349

Service pages126

Source: Google Search Console, page-level data for the 33 new pages published 3 to 5 September (21 and 22 September not final when pulled). The company facts page, published 22 September, had no impressions.

When inspected through Google's URL Inspection API on 22 September, 33 of the 34 new pages were indexed. 32 of them drew their first Google impression on the day they went live or the next day, and the company facts page was indexed on the day it was published. The one page left out is a service page. Google's last crawl of it came minutes after it went live on 3 September and recorded a noindex directive in the HTTP header.

From 3 to 22 September the new pages drew 5,715 Google impressions and 20 clicks. Comparison listicles drew 4,615 of those impressions (80.8%) and 14 of the clicks, at an average position of 27.3. Industry pages drew 625 impressions at an average position of 43.3. The four guide and research pages drew 349 impressions at an average position of 7.7 (page-types.csv).

Conversational queries found the listicles near the top

Search Console shows the query text behind 3,181 of the new pages' 5,715 impressions from 3 to 22 September (55.7%). Of those visible impressions, 669 (21.0%) came from 145 long conversational queries, defined as 9 or more words or ending in a question mark. They averaged position 6.9. Keyword queries of 8 words or fewer, 2,512 impressions, averaged position 51.0. Neither group produced a click.

Comparison listicles drew 641 of the 669 conversational impressions. The query with the most impressions was "which lead generation companies work with manufacturers and long sales cycles?" (31 impressions, average position 1.3). Six queries, with 31 impressions between them, were a pasted instruction to an AI model, such as "answer the following user question naturally and concisely. if relevant, include specific tools/companies and explain why. question: what is the best fully managed outbound sales agency for manufacturing companies?" (22 impressions, average position 2.0). Search Console does not say who ran these searches. Their shape matches the web searches that AI assistants and AI-visibility tools run while they compose an answer (conversational-query-examples.csv).

Google clicks rose, mostly on older pages

Chart data

date	Google clicks
2026-07-01	24
2026-07-02	21
2026-07-03	13
2026-07-04	6
2026-07-05	16
2026-07-06	21
2026-07-07	12
2026-07-08	26
2026-07-09	12
2026-07-10	14
2026-07-11	9
2026-07-12	7
2026-07-13	11
2026-07-14	12
2026-07-15	16
2026-07-16	14
2026-07-17	13
2026-07-18	7
2026-07-19	4
2026-07-20	9
2026-07-21	11
2026-07-22	11
2026-07-23	8
2026-07-24	10
2026-07-25	3
2026-07-26	11
2026-07-27	10
2026-07-28	17
2026-07-29	9
2026-07-30	11
2026-07-31	15
2026-08-01	1
2026-08-02	10
2026-08-03	9
2026-08-04	8
2026-08-05	9
2026-08-06	15
2026-08-07	15
2026-08-08	2
2026-08-09	3
2026-08-10	7
2026-08-11	14
2026-08-12	17
2026-08-13	11
2026-08-14	7
2026-08-15	11
2026-08-16	8
2026-08-17	10
2026-08-18	11
2026-08-19	12
2026-08-20	10
2026-08-21	12
2026-08-22	5
2026-08-23	5
2026-08-24	16
2026-08-25	13
2026-08-26	8
2026-08-27	6
2026-08-28	11
2026-08-29	6
2026-08-30	9
2026-08-31	12
2026-09-01	7
2026-09-02	10
2026-09-03	10
2026-09-04	11
2026-09-05	3
2026-09-06	9
2026-09-07	13
2026-09-08	18
2026-09-09	12
2026-09-10	18
2026-09-11	17
2026-09-12	10
2026-09-13	7
2026-09-14	10
2026-09-15	16
2026-09-16	16
2026-09-17	16
2026-09-18	18
2026-09-19	8
2026-09-20	10

Source: Google Search Console, domain property, final data only. The new pages went live on 3 September, the same day as the changes to older pages listed in changes.csv.

Across the whole site, Google clicks rose 41.0% (134 to 189) and impressions 44.1% (12,796 to 18,440) between the windows of 17 to 30 August and 7 to 20 September. Average position went from 26.0 to 19.9. The 7-day check gives 36.2% more clicks (69 to 94). The before window also fell in a late-August dip: against 6 to 19 July, another Monday-to-Sunday window, clicks from 7 to 20 September were 6.2% higher (178 to 189).

Page-level data split the 54 extra clicks. The 34 new pages supplied 16. Older product pages went from 50 to 99 clicks, while other older pages fell from 85 to 74. 28 older pages that had no Google impressions at all from 1 to 30 August drew 12 clicks in the after window. On 3 September, the day the new pages went live, 15 of the older product pages got new titles, and all of them lost their unsupported claims and moved to one URL each. The rise in Google clicks cannot be credited to the new pages.

What showed no gain

• Category answers. "best B2B lead generation agencies" and "best cold email agencies for B2B" named the company in 0 of 10 answers across ChatGPT, Perplexity, Gemini, Claude and Grok on 22 September. The tracker's 10 generic prompts named it in 2 of 40 answers that day, both from ChatGPT. Branded prompts named it in 25 of 25 answers because the prompt contains the name (ai-scan.csv).
• Brand search. Google clicks on searches for the company's current or former name were 24 from 17 to 30 August and 31 from 7 to 20 September, and impressions fell from 412 to 362. The two weekly values after the launch, 15 and 16, sit inside the range of 10 to 31 recorded in the 13 weeks from 1 June to 30 August (branded-search-weekly.csv).
• Bing. The 34 new pages drew 26 Bing impressions and no clicks from launch through the week of 18 September, all on 6 of the pages. Whole-site Bing clicks went from 11 to 10 between the matched windows, and impressions rose 25.1% (854 to 1,068).

ChatGPT's API often answered without searching

On 22 September we asked ChatGPT 10 prompts through the OpenAI API with its web search tool enabled, then asked again 3 of the 4 prompts whose answers came back without citations. 7 of the 13 answers had no citations, and for the 3 repeat questions the API reported that it ran no web search. Both answers to "Who founded [company] and when?" named a founder the company never had, a different one each time, and gave 2020 as the founding year. Neither answer to "Is [company] legit?" gave a verdict. The daily tracker used the same API model, so its ChatGPT counts include answers written without a search. The consumer ChatGPT app may search more often than the API did here.

Method

One site. This is a study of one site: a US done-for-you lead-generation agency founded in 2022. It covers 3 to 22 September 2026, with baselines from June to August. It shows what happened on this site and says nothing about what the same pages would do elsewhere.

Several changes landed together. On 3 September the site published 32 pages and, the same day, noindexed 164 template pages, merged 8 duplicates, retitled 15 older product pages, removed unsupported claims from the template behind its product pages and redirected alternate URLs (changes.csv). It added links to other websites on 16 existing pages on 19 and 20 September and published a company facts page on 22 September. Any before-and-after difference mixes these effects. We credit a result to the new pages only where it is measured on those pages: ChatGPT visitors by entry page, AI citations by cited URL, and Search Console data by page.

How the windows were matched. The main comparison uses two 14-day windows that each run Monday to Sunday, so each holds exactly two of every weekday: 17 to 30 August 2026 before and 7 to 20 September 2026 after, 21 days apart. The launch week, 31 August to 6 September, is in neither window. The after window ends on 20 September, the last day of final Search Console data when the data were pulled on 22 September. A 7-day check compares 24 to 30 August with 14 to 20 September, and a context window runs from 6 to 19 July, both also Monday to Sunday. The same windows are used for Search Console, Plausible and Bing.

Sources.

• Google Search Console, domain property: daily whole-site totals (final data only), clicks and impressions by page and date, and by query and page for the new pages. Page and query figures for 3 to 22 September include 21 and 22 September, which were not final when pulled.
• Google URL Inspection API, run on 22 September for all 34 new pages.
• Plausible analytics (Stats API): visitors by referral source, entry pages of ChatGPT visitors, and a goal for the page shown after a contact form is sent. Window totals are unique visitors; weekly values in Exhibit 1 are sums of daily visitors. Plausible counts a visit as ChatGPT when it carries a ChatGPT referrer or tag, so visits from ChatGPT without one count as direct and the ChatGPT figures are a floor. Figures for 22 September cover part of that day.
• Bing Webmaster Tools API: daily whole-site traffic and weekly page statistics through the week of 18 September.
• The daily tracker's own records: 43 runs from 27 July to 9 September 2026, each asking every prompt once per engine and storing every cited URL. Models: OpenAI chat-latest with web search (ChatGPT), Perplexity sonar, gemini-2.5-flash (Gemini) and, from 29 July, claude-sonnet-5 with web search (Claude). Engine errors left 34 of 1,540 expected answers missing before 6 September and 20 of 160 from 6 to 9 September, and ChatGPT returned no answers on 4 September. Missing answers are left out of every count.
• A scan on 22 September through each engine's API with web search on: the tracker's 10 prompts on its 4 engines, and 10 more prompts on 5 engines, adding Grok (grok-4.7). One answer per prompt per engine, plus 3 repeat questions to ChatGPT (prompts.csv).
• A review of the live new pages on 22 September, for the page descriptions and the listicles' self-rankings.
• The site's change history, for the dates and scope of each change (changes.csv).

Definitions. A page is new if it was first published from 3 to 22 September; one older listicle rewritten on 3 September counts as an older page. Named: the company's current or former name appears in the answer. Cited: a URL on the company's domain appears among the answer's sources or links. Conversational query: 9 or more words, or ending in a question mark. Keyword query: every other visible query. Older product pages: the site's pages for the contact data it sells, one per profession, identified by URL pattern. Indexed within a day: a first Google impression on the day of publication or the next day, or, for the page published on 22 September, indexed status when inspected that day. Brand-name searches: 8 query variants that are the company's current or former name, leaving out an unrelated company with a similar name and lookups of the domain's email settings.

AI answers vary. An engine can answer the same prompt differently from one run to the next and from one day to the next. The tracker asked each prompt once per engine per run, and the 22 September scan is a single draw. Counts such as 4 of 40 or 0 of 10 are small samples, and a repeat on another day could differ.

Limits. One site, one launch and three weeks. Referral and conversion counts are small. Search Console withheld the query text for 2,534 of the new pages' 5,715 impressions (44.3%), so the query-type split covers the visible 3,181 only. The study measures traffic and citations; it makes no judgment on the accuracy of the pages' content, and an AI citation is no evidence of accuracy.

Data

The CSV files below hold every number in this report. Page addresses, the company's name, its prices and competitor names are removed; prompts show placeholders such as [company]. The files are free to reuse under CC BY 4.0 with credit to Strandline Research.

• Daily Google traffic and ChatGPT visitors CSV, 82 rows, 2.1 KBOne row per day, 1 July to 20 September 2026: whole-site Google clicks, impressions and average position (final data), and ChatGPT referral visitors from Plausible.
• ChatGPT referral visitors by week CSV, 16 rows, 0.4 KBSum of daily ChatGPT referral visitors for each Monday-to-Sunday week from 1 June to 20 September 2026.
• Before and after windows CSV, 17 rows, 2.1 KBEvery before-and-after comparison in the report: metric, source, window dates, before and after values, change and percentage change. Main windows are 17 to 30 August and 7 to 20 September 2026.
• The 34 new pages by type CSV, 6 rows, 1.0 KBBy page type: pages, publication dates, pages indexed on 22 September, pages with a first Google impression within a day of publication, pages indexed within a day, Google clicks, impressions and average position (3 to 22 September and 7 to 20 September), visible and conversational query impressions, Bing pages, impressions and clicks through the week of 18 September, and ChatGPT entry visitors (3 to 22 September).
• Google impressions by type of new page CSV, 4 rows, 0.2 KBGoogle impressions and clicks from 3 to 22 September 2026 for the four page types published 3 to 5 September, with impressions per page. Source of Exhibit 3.
• Visible Google queries on the new pages by type CSV, 6 rows, 0.5 KBDistinct queries, query and page pairs, impressions, share of visible impressions, clicks and average position for conversational and keyword queries on the 34 new pages, 3 to 22 September 2026.
• Examples of conversational queries CSV, 15 rows, 2.5 KBThe 10 conversational queries with the most impressions on the new pages, plus every query containing a pasted AI instruction, with the page type, impressions and average position. Queries naming companies are left out.
• Where the extra Google clicks came from CSV, 7 rows, 0.6 KBPages with impressions, clicks and impressions before (17 to 30 August) and after (7 to 20 September) for the new pages, older pages already in Google results, older pages with no August impressions, and the noindexed template pages, with older pages also split into product pages and other pages.
• Entry pages of ChatGPT visitors CSV, 10 rows, 0.3 KBChatGPT referral visitors by the type of page they entered on, 1 June to 2 September and 3 to 22 September 2026.
• Daily AI tracker runs CSV, 44 rows, 2.1 KBOne row per run of a daily tracker that asked 10 buyer prompts to up to 4 AI engines, 27 July to 9 September 2026, plus the re-run of 22 September: answers expected, returned and missing, answers naming the company, citing its site and citing a new comparison listicle, for all engines and for ChatGPT.
• AI answers by prompt type and engine, 22 September 2026 CSV, 24 rows, 1.3 KBFor each prompt type and engine: prompts, answers, answers naming the company and answers citing its site.
• ChatGPT API answers, 22 September 2026 CSV, 13 rows, 1.7 KBThe 13 ChatGPT answers from the 22 September scan: pass, prompt type, prompt with names replaced, number of citations, web searches reported by the API (re-asks only), whether the answer names the company, and what the answer did.
• Prompts CSV, 20 rows, 3.2 KBAll 20 prompts used, with the company and competitor names replaced by placeholders, and the engines each was asked on.
• Brand-name searches by week CSV, 24 rows, 0.6 KBGoogle clicks, impressions and average position for searches of the company's current or former name, by Monday-to-Sunday week from 6 April to 20 September 2026.
• Where the publisher ranked itself CSV, 9 rows, 0.1 KBThe 12 new comparison listicles grouped by the number of vendors listed and the publisher's own position.
• What the site changed, by date CSV, 11 rows, 0.9 KBDated list of the changes the site made from 3 to 22 September 2026, with the number of pages each change touched where it applies to a set of pages.