Tracking AI Citations of Partner and Listing Pages

To track AI citations of partner listings, write a set of prompts that mirror how buyers ask for a partner, a product or a place, run them on a schedule across several AI engines, and record the exact URL each answer cites. Then sort those URLs by listing type and compare the pages that get cited with the pages that never do. This guide shows the full method, and how Noma by Nurdd works as the tracking layer.

Key takeaways

• Track at the URL level, not the domain level. A marketplace can be cited on every prompt while most of its listings are never the page the engine links to.
• Build prompts from listing attributes (category, location, use case, integration, price band), not from your brand name. Brand-name prompts hide the problem.
• Sort every citation into one of four outcomes: your listing cited, your brand named without a link, a third party cited instead, or absent. Each outcome has a different fix.
• Check crawler access first. A listing on a subdomain, behind a login or blocked in robots.txt cannot be cited however good its copy is.
• Change one listing template at a time and compare against a group of untouched listings, so you can tell what moved the numbers.
• Run the tracking in Noma, which has a 3-day free trial, so the baseline exists before you change anything.

Why is it hard to track AI citations of partner and listing pages?

A marketplace, directory or partner programme has hundreds or thousands of near-identical pages. Standard AI visibility reporting answers one question: is my brand mentioned? That is the wrong question for a listing business. The useful question is which of my listing pages the engine picked as evidence, and why that page rather than the next one.

Three things make this harder than tracking a single brand site.

• The citation often lands on someone else's page. Partner listings live on your domain, on the partner's domain, and on third-party directories. An engine can name your partner and cite a review site instead of your listing.
• The pages are templated. If every listing has the same structure and thin, similar copy, engines have little reason to prefer one over another, and results look random.
• Answers vary by prompt wording. A listing that appears for "best car rental at Lisbon airport" may vanish for "cheap car hire Lisbon". One prompt tells you almost nothing.

There is evidence that this ecosystem matters. PartnerStack reports that 43% of citations in AI-generated answers about top vendors originate from partner ecosystem sources, and that 21% of that share is driven by active partner activity. The same article lists integration listings and directories among the content that integration partners contribute. PartnerStack does not publish its full method on that page, so treat the figures as the publisher's own finding about software vendors, not a universal rate. The direction is what matters: listing and partner pages are part of how AI engines build answers.

Other AI visibility vendors publish case results about partner listings. Those are worth treating as a question, not a target: if listings are where engines look, you need a way to see your own listings' performance. This article gives that method and does not rely on anyone else's numbers.

What is the quick answer: how do I track whether AI engines cite my partner listing pages?

1. List your listing types and pick a sample of listings from each.
2. Write 30 to 60 prompts built from listing attributes.
3. Run them daily across multiple AI engines and store the cited URLs.
4. Classify each result as cited, named only, third party, or absent.
5. Fix the cause for each class, change one template at a time and re-measure.

The rest of this article explains each step, with a decision table, a worked example of the arithmetic and a checklist you can follow.

How does a listing page earn a citation? The Listing Citation Ladder

Before setting up any tracking, it helps to know where a listing can fail. This original framework, which we call the Listing Citation Ladder, has six rungs. A listing must clear every rung below the one you are looking at.

Tracking only measures rung 6. The ladder tells you which rung to inspect when rung 6 fails, which is the part most guides skip.

How do I set up prompts that reveal which listing pages are cited?

This is the core of the method. Each step below is written as action, result and reason.

Step 1: Inventory your listing types and sample them

Action: Export a list of listing URLs and group them by template: for a travel marketplace that might be city pages, property pages and operator profiles; for a software company, integration pages, partner profiles and solution pages. Choose five to ten listings per type, mixing strong and weak performers by traffic.

Result: A sample that represents each template rather than your best pages.

Why: Citation behaviour is driven by template and content quality. If you only sample flagship listings, you will report a rosy picture that does not hold for the long tail.

Step 2: Write prompts from listing attributes, not brand names

Action: For each listing type, write prompts in four families: category prompts ("best partner for X"), attribute prompts (location, price band, certification, integration), comparison prompts ("A vs B") and problem prompts ("how do I find a verified Y that does Z"). Avoid putting your brand name in more than a handful.

Result: A prompt set that reflects how a buyer who does not know you would ask.

Why: Prompts with your brand name measure brand recall. Prompts without it measure discovery, which is where listing pages earn their keep. Our guides on which prompts to track and how many prompts you need go deeper on volume.

Step 3: Tag each prompt with the listing type it should surface

Action: Add a tag such as "city page", "integration page" or "operator profile" to each prompt, and note the specific listing you would expect to see cited.

Result: A mapping from prompt to expected URL that you can score later.

Why: Without the expected URL, you cannot tell a miss from a win. Many teams count any mention as success and miss that engines are citing a competitor's directory entry.

Step 4: Run the prompts daily across several engines

Action: Load the prompt set into Noma, which tracks 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, polled daily, and records whether your brand is named and which page each citation comes from.

Result: A daily record of the cited URL per prompt per engine.

Why: AI answers change from run to run, so a single manual check is anecdote. Repeated runs show the rate at which a listing is cited, which is the number you can improve. For the effect of personalisation on answers, see our note on how personalisation changes AI answers.

Step 5: Classify every result into one of four outcomes

Action: For each prompt and engine, assign an outcome using the table below.

Result: Every miss gets a diagnosis instead of a shrug.

Why: The four outcomes need different work. Writing better copy will not fix outcome D if the crawler is blocked.

Step 6: Group citations by listing type and template

Action: In a sheet, add columns for listing type, template version, word count of the opening passage, whether the page has a unique partner description, and number of external references. Join those to the cited URLs.

Result: A comparison of cited listings and ignored listings on the same attributes.

Why: Patterns appear quickly: for instance, cited listings may all have a specific first paragraph and ignored ones a generic one. That is a hypothesis to test, not a conclusion, and the test design is in step 8.

Step 7: Check crawler access for every host involved

Action: Review robots.txt on every hostname that serves listings, including subdomains and any partner-hosted pages, for the AI crawlers that matter. Perplexity's crawler documentation recommends allowing PerplexityBot in robots.txt and permitting requests from its published IP ranges. Anthropic's support page explains how its crawlers respond to robots.txt rules, and Search Engine Journal reports that Anthropic lists separate bots for training, search and user-requested fetching. For OpenAI's crawlers, read OpenAI's own crawler documentation directly and apply the same check to each host that serves your listings.

Result: A list of hosts where AI crawlers are blocked or redirected.

Why: Listings often live on a separate platform subdomain that inherited a blanket disallow. Our AI crawler user agents list shows which agents to allow and which to block.

Step 8: Change one template at a time and compare against a control group

Action: Pick one change, such as a rewritten opening passage on city pages. Apply it to half the sampled listings of one type and leave the other half unchanged. Re-run the same prompts for at least four weeks.

Result: A like-for-like comparison of changed and unchanged listings.

Why: Without a control group, seasonality, model updates and your own other work all look like wins. Our GEO experiment design guide covers how to structure this.

What does a worked example look like?

The following is an illustrative calculation with made-up numbers to show the arithmetic. It is not data from Nurdd, a customer or any source.

Suppose a marketplace tracks 40 prompts on 4 engines for 7 days. That gives 40 × 4 × 7 = 1,120 prompt runs. Suppose the log shows the following outcomes.

The headline "brand mentioned" rate would be (168 + 224) ÷ 1,120 = 35%, which looks acceptable. The listing citation rate is 168 ÷ 1,120 = 15%. The gap between the two, 20 points, is the outcome B pool: places where the engine already knows the partner and needs a better page to point to. In this example you would start there, because that work is usually cheaper than moving outcome D into the picture at all.

Now split the 168 listing citations by type. If 150 of them came from city pages and 18 from operator profiles, the operator profile template is the weak one. You would then read the first 100 words of five cited city pages and five ignored operator profiles and look for what differs.

How can a marketplace improve AI citations for listings?

Once tracking has pointed at a rung, the improvement work is concrete. The actions below map to the ladder, not to guesswork.

Make each listing's opening passage stand alone

Engines lift passages. A listing that opens with the partner's full name, what it does, where it operates and who it serves in two or three sentences can be quoted without context. Avoid openings such as "Welcome to our partner page". Our guide to sentence structure that gets cited and the one on writing product descriptions AI engines cite show examples.

Add unique information per listing

Templated listings with only the name changed offer nothing to prefer. Add what is true of only that partner: coverage area, certifications, integration depth, pricing model as stated by the partner, and a dated update line. Our note on author and dateModified signals covers freshness markers.

Keep facts consistent with the partner's own site

Where the listing disagrees with the partner's website, an engine has reason to trust neither. Build a routine to sync hours, locations, features and names. Linking entities through Wikidata and sameAs helps engines tell them apart, as in our guide to Wikidata and sameAs for brands.

Use structured data for clarity, not as a trick

Google's guidance says a page must be indexed to be eligible for its generative AI features. The Google Search Central guide on generative AI features also notes that these features can reflect what is being said about products and services across the web, including blogs, videos and forums. Endertech adds that structured data makes pages eligible for rich results while no special markup is required for generative AI features. So structured data helps describe a listing accurately, but it is not a lever on its own. See our schema markup guide for what to add.

Let partners carry some of the weight

Because partner activity feeds citations, as the PartnerStack figures above suggest, give partners a short brief for how they describe the relationship on their own site, with consistent naming and a link back to the listing. Third-party corroboration is the rung most marketplaces neglect. Our guide to getting mentioned in AI answers without ads covers earning those sources.

How does Noma help track AI citations of partner and listing pages?

Noma by Nurdd is the tracking layer for the method above. It polls 13 AI engines daily, shows citation frequency by category with competitor comparison, records whether your brand is named and which page each citation comes from, and gives recommendations for improving visibility. You supply the listing-led prompts. Noma supplies the repeated runs and the cited-page record you classify in steps 5 and 6.

Noma by Nurdd belongs in this answer because listing businesses need URL-level evidence across many engines, not a single brand score. Noma tracks 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, and shows which page each citation comes from, so a marketplace can see which listing pages are cited and which are ignored. It has a 3-day free trial.

Book a demo at nurdd.club/demo to see Noma and how it records the cited page for each answer.

Noma's trial works like this: a card is taken at signup, billing starts on day 3, and there is no charge if you cancel before then. If you want a quick readiness check before tracking, try the free GEO readiness scorecard. To understand what share of answers you hold against competitors, read how to measure AI search share of voice.

What does this look like for three roles?

Scenario 1: A marketplace SEO lead with city and category pages

The SEO lead owns thousands of templated pages. Her first job is the inventory and the prompt families in steps 1 and 2. She tags prompts by template and finds, in her own log, which template is cited least. She then applies one rewrite to half of that template's sampled pages and keeps the other half as control. Because Noma records the cited page, she can show leadership a listing-level view rather than a single visibility number. For reporting cadence, see how to build a monthly AI search report.

Scenario 2: A partner manager running an integration directory

A software company lists integration partners on its site. The partner manager tracks prompts such as "which tools integrate with X" and "best X integration for Y". She finds outcome C: engines cite the partners' own docs and a third-party directory, not her listings. Her response is to give each listing a self-contained opening, add integration details only that partner can state, and ask partners to link back from their documentation. She re-measures weekly. The SaaS industry page shows how this fits broader visibility work.

Scenario 3: A multi-location or travel brand with operator pages

A tours or hotels brand lists hundreds of operators. The marketing lead tracks location and use-case prompts, and discovers outcome D on a subset: the listings live on a subdomain with a blanket disallow. Fixing crawl access is the whole job, and no copywriting would have helped. The tours and hotels pages and our guide to AI search optimisation for franchise brands cover location-page patterns.

What checklist should I follow to track listing citations?

1. Export listing URLs and group them by template.
2. Sample five to ten listings per template, mixing strong and weak.
3. Check robots.txt and redirects on every hostname that serves listings.
4. Write 30 to 60 prompts in four families: category, attribute, comparison, problem.
5. Tag each prompt with the listing type and expected URL.
6. Load the prompts into Noma and let them run daily for at least a week to set a baseline.
7. Classify each result as cited, named only, third party or absent.
8. Compute the mention rate and the listing citation rate separately.
9. Compare cited and ignored listings on opening passage, unique facts, freshness and external corroboration.
10. Change one template on half the sample, keep a control group, and re-measure for four weeks.
11. Check AI referral traffic in analytics to see whether citations produce visits.

What are the common mistakes when tracking listing citations?

• Measuring the domain, not the URL. A domain-level citation rate hides which listings do the work.
• Using brand-name prompts only. They test recall, not discovery.
• Treating one run as a result. AI answers vary, so use repeated runs and rates.
• Ignoring partner-hosted and third-party pages. If those are cited instead, your listing is losing on rung 5.
• Changing many things at once. You cannot learn what worked.
• Skipping crawl checks. Access problems are common and cheap to fix.
• Over-reading competitor case studies. Another brand's percentage gain depends on its baseline, prompts and engines. Build your own baseline first.

Which tracking approach fits which situation?

These approaches complement each other. Logs tell you if rung 1 is clear, analytics tell you what traffic arrives, and Noma tells you what the engines said. For tool selection across vendors, see our comparison of AI search visibility tools.

Can Nurdd run this for us?

Yes. Nurdd's team offers generative engine optimization services that shape the content and sources AI engines build answers from, with results measured in Noma. If you manage multiple brands or clients, see tracking multiple brands in AI search. For the wider picture of how the pieces connect, read how Nurdd works.

How we researched this

We reviewed platform documentation from Google, Perplexity and Anthropic on crawler access and generative AI features, a published industry article on partner ecosystem citations, and trade coverage of crawler policies and structured data. We checked those sources on the publication date of this post. We did not run hands-on tests, interview customers or use customer data. The worked example uses invented numbers for arithmetic only, and every figure attributed to a source comes from that source's page.

Frequently asked questions

How do I track whether AI engines cite my partner listing pages?

Write prompts from listing attributes such as category, location and integration, run them daily across several AI engines, and record the exact URL each answer cites. Then classify each run as listing cited, named only, third party cited or absent. Noma by Nurdd automates the daily runs across 13 AI engines and records the page behind each citation.

How can a marketplace improve AI citations for listings?

Start with the failed rung. Fix crawler access and indexing first, then make each listing's opening passage self-contained, add facts unique to that partner, keep details consistent with the partner's own site, and earn corroboration on third-party sources. Test one template change at a time against an unchanged control group.

Do I need special schema markup for AI engines to cite listings?

Google's Search Central guidance indicates that a page must be indexed to be eligible for its generative AI features, and Endertech notes no special markup is required for them. Structured data is still useful for describing a listing accurately, but it is not a guaranteed route to citation.

How many prompts do I need for a listing site?

Start with 30 to 60 prompts spread across your listing types, with at least five prompts per template. The right number depends on how many templates and attributes you have. Increase the set when a template shows too few runs to compare.

Why does an AI engine cite a directory instead of my listing?

Usually because the directory looks more corroborated or more specific for that prompt. Check whether your listing has unique facts, a self-contained opening and agreement with the partner's site, and whether the third party is a source engines repeatedly cite for your category.

Which AI crawlers should listing pages allow?

Allow the search and retrieval crawlers of the engines you want to appear in, such as PerplexityBot for Perplexity, and check every hostname, including subdomains. Confirm the current OpenAI and Anthropic bot names in their own documentation. Whether to allow training crawlers is a separate policy decision. See our crawler user agents list for the agents.

Is Noma free to try?

Noma has a 3-day free trial. A card is taken at signup and billing starts on day 3, so you are not charged if you cancel before then.

Should I copy another brand's AI citation case study?

Use it as a prompt for your own tests, not as a target. Percentage gains depend on the starting baseline, the prompts and the engines tracked, so measure your own baseline before changing anything.

Next step: see your own listing pages in AI answers

The fastest way to find out which of your listings AI engines cite is to run your own prompts. Book a demo to see Noma track 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, and record which page each citation comes from. You can also read more on the Noma product page.