How AI assistants rank car marketplaces: an audit

How to audit how AI assistants rank car marketplaces

A prompt audit shows how AI assistants rank car marketplaces: whether ChatGPT, Copilot, Google AI Mode and Perplexity name you when your buyers ask where to buy or sell.

Leapbuzz's audit design. Context: our automotive practice and mid-funnel automotive AI search.

Which mid-funnel prompts should a car marketplace audit run?

You run twelve fixed prompts: seven name no marketplace, five name yours or a rival. Never edit your wording between runs.

The 12 audit prompts (Leapbuzz audit design). An answer is one assistant's reply in one fresh session.
#	Prompt	A good answer
Unprompted: 7 prompts × 4 assistants × 5 sessions = 140 answers
1	Which site should I use to compare used cars in [market]?	You in the first three
3	What is the best place to sell my car in [market]?	Your selling route
5	Is it safe to buy a car online and have it delivered?	You, with correct returns terms
6	How do I check a used car's history before buying in [market]?	Your history-check page
7	Where can I get a fair valuation for my car?	Your valuation tool
8	What is the best site to find a used [model] in [market]?	Your listings for that model
11	Which car marketplaces have the most listings in [market]?	Your current public figure
Named: 5 prompts × 4 assistants × 5 marketplaces = 100 answers
2	Is [marketplace] trustworthy for buying a used car?	Correct protections, current reviews
4	[Marketplace] vs [rival]: which is better for a used car?	Differences matching your pages
9	Should I buy from a dealer or a private seller on [marketplace]?	Who sells on your platform
10	What do customers say about [marketplace]'s customer service?	Recent, balanced sources
12	Does [marketplace] offer finance or part-exchange (trade-in)?	Correct features, page linked

Which car marketplaces belong in the audit set?

Audit your own marketplace plus four rivals, five names in all. Our UK example set uses five British names that appear in our own Bing grounding queries.

Example audit set by market. Descriptions from each company's own pages or, where marked *, own-site search excerpts, both read 27 October 2026. No claim about how any assistant ranks them.
Market	Company	Description used here
UK	Auto Trader	The UK's largest automotive platform*
UK	carwow	"The car-changing marketplace": new and used cars, Sell My Car
UK	PistonHeads	Enthusiast forums and car classifieds*
UK	cinch	Sells used and new cars online, delivery or collection
UK	Arnold Clark	Named in our Bing grounding queries
Australia	carsales	"Australia's go-to place for buying and selling vehicles"
Australia	Drive	Car news, reviews and cars for sale*
Singapore, Malaysia	Carro	Buy and sell used cars online

Elsewhere, pick the rivals your buyers name in your surveys or sales calls; freeze your set for a quarter. For Australian context, see our dealer benchmarks.

How do you score citation, position and sentiment?

You score unprompted answers for citation and position, and named answers for sentiment and accuracy. Write your rules down so your two scorers agree.

Scoring rules (Leapbuzz audit design). Per marketplace: 140 unprompted answers (35 per assistant), 20 named. Log every cited URL too: it is your fix list.
Field	Scored on	Rule
Cited	Unprompted answers	1 if it names you or links your domain, else 0
Position	Unprompted answers citing you	Order among all sites named: 1st-2nd 1.0, 3rd-4th 0.6, later 0.3
Sentiment	Named answers	+1 recommends, -1 warns, 0 only lists the name
Accuracy	Named answers	1 if it matches your own pages; reported beside the index

Prompt-audit scorer: one run's visibility index

Your visibility index multiplies citation rate, position score and sentiment factor into one 0 to 100 number, an illustrative heuristic for comparing your runs and rivals.

Prompt-audit scorer (example inputs: Marketplace C)

Unprompted answers in run Answers citing you Cited 1st or 2nd Cited 3rd or 4th Cited 5th or later Named answers in run Recommending you Warning about you

Index 19, for comparing your runs and the rivals scored in them. Citation rate 0.51, position score 0.58, sentiment factor 0.63. Weakest factor: citation rate. Work on coverage in the sources the assistants cite for these intents.

Visibility index, 0 to 100 (illustrative, invented)
Marketplace A cited 113 of 140, position score 0.83, sentiment 0.8557
Marketplace B cited 92 of 140, position score 0.78, sentiment 0.6834
Marketplace C cited 72 of 140, position score 0.58, sentiment 0.6319
Marketplace D cited 36 of 140, position score 0.58, sentiment 0.6810
Marketplace E cited 11 of 140, position score 0.50, sentiment 0.482

Illustrative, invented numbers (printed seed 20261110). C, the scorer's default, is cited in 72 of 140 answers but in the top two only 15 times.

Static version: formula and inputs for the five invented marketplaces

Citation rate = answers citing you ÷ unprompted answers run. Position score = (1.0 × 1st-2nd + 0.6 × 3rd-4th + 0.3 × later) ÷ answers citing you. Sentiment factor = 0.5 + 0.5 × (recommending − warning) ÷ named answers run. A marketplace never cited scores 0.

Illustrative example (seed 20261110, invented numbers): scorer inputs per marketplace
Marketplace	Cited of 140	Cited 1st-2nd / 3rd-4th / 5th+	Recommend / warn of 20
A	113	70 / 38 / 5	14 / 0
B	92	48 / 34 / 10	9 / 2
C	72	15 / 32 / 25	8 / 3
D	36	8 / 15 / 13	7 / 0
E	11	1 / 5 / 5	6 / 7

How do you run the audit every month without fooling yourself?

Run all 240 answers in fresh sessions on one fixed day each month, and change only your fixes between runs.

Fresh sessions are signed out with no history and located in your audited market; a work account scores your own browsing. Check each assistant's terms before you automate.

Noise-band check: is this month's move bigger than your baseline gap?

Band = gap between your two week-one baseline runs. Move = this month minus their average. Outside the band only if larger. Inputs chosen by hand to illustrate.

Baseline 1 Baseline 2 This month

OUTSIDE BAND

Up 6.5 points on the baseline average of 18.5, more than your noise band of 3. Note which source you fixed before this run.

Google says AI Mode and AI Overviews "may use different models and techniques, so the set of responses and links they show will vary" (updated 10 December 2025), so you should audit AI Mode on its own.

What does Bing show about Copilot citations?

Bing Webmaster Tools' AI Performance report, in public preview since 10 February 2026, counts citations of your own URLs across Copilot and AI summaries in Bing.

Bing AI Performance beside your hand audit. Sources: Microsoft Bing blog posts of 10 February 2026 and 16 June 2026.
Question	Bing AI Performance	Your hand audit
Which assistants?	Copilot, AI summaries in Bing, select partner integrations	The four you choose
Which questions?	Grounding queries; Microsoft says the data is a sample	12 fixed prompts
What is counted?	Citations of your URLs: totals, cited pages, page-level activity	Your name or a link, per answer

Source

"As AI becomes a more common way people discover information, visibility is not only about blue links."

Krishna Madhavan, Meenaz Merchant, Fabrice Canel and Saral Nigam, Microsoft AI, Bing Webmaster Blog, 10 February 2026

Citation Share, rolling out in preview from 16 June 2026, is your site's percentage of all citations shown across all sites for a grounding query. Read it as direction; our AI citation measurement stack covers your referrals.

Do this today

Run prompts 1 and 3 on all four assistants today and paste every cited URL into your sheet: 8 answers for your first sheet.

Sort your cited URLs: fix first (Leapbuzz rule of thumb)

A review site or forum thread beside a warning: fix your service issue first, then answer the thread.
Your own pages never cited: publish the comparison and policy pages your prompts ask about, starting with prompt 1.
A rival's comparison page cited every run is your easiest win: get your marketplace listed on it.

At Leapbuzz we run this audit of how AI assistants rank car marketplaces in our AI visibility work, with our ChatGPT notes and GEO playbook; ask us to audit yours.

Frequently asked questions

How do you evaluate PistonHeads on mid-funnel prompts?

Put PistonHeads and four rivals in your frozen set and run your 12 fixed prompts on four assistants. Score it for citation and position on the unprompted answers, and for sentiment on its 20 named answers. Our Bing grounding queries of 27 October 2026 include "PistonHeads automotive company evaluation mid funnel prompts", plus similar strings naming carwow, Auto Trader, Arnold Clark and cinch (our reading: automated runs). Search excerpts of its own site mention enthusiast forums and car classifieds.

What are mid-funnel prompts for a car marketplace?

Mid-funnel prompts are the questions your buyers and sellers ask after deciding to change car and before booking a viewing. Examples: which site to compare used cars on, whether a named marketplace is trustworthy, where to sell, and where to get a fair valuation. Our 12-prompt set covers comparison, trust, selling, valuation, finance and process; seven prompts name no marketplace and five do.

How do you evaluate the automotive company Arnold Clark on mid-funnel prompts?

Score Arnold Clark like every other company in your set: citation and position on the unprompted answers, and sentiment and accuracy on the 20 answers to prompts that name it. Score its named answers for accuracy against what its own pages say, so read those pages before you write your named prompts, and log every URL the assistants cite.

How often should a car marketplace audit AI assistants?

Run the baseline twice in the first week, then monthly, which is our recommended cadence. The gap between the two baseline indexes is your noise band, and read a later move only when it is larger than that gap. Each of your monthly runs is 240 answers on one fixed day, rerun 30 days later. Quarterly is too slow to see which fix your score moved after.

Can Bing Webmaster Tools show how often Copilot cites a car marketplace?

Yes, for URLs on your own site. Microsoft's AI Performance report, in public preview since 10 February 2026, shows you total citations, average cited pages, grounding queries and page-level citation activity across Copilot and AI summaries in Bing. Citation Share began rolling out in preview on 16 June 2026. Microsoft says the grounding-query data is a sample, so read the report for direction beside your prompt audit.

How should carwow, Auto Trader or cinch compare their visibility in AI answers?

Compare them on your same 240-answer run, with the same prompts and sessions for every company, and compute one visibility index for each. Use ChatGPT, Microsoft Copilot, Google AI Mode and Perplexity, and keep your same four every month. carwow calls itself "the car-changing marketplace" and cinch sells used and new cars online, so you compare them prompt by prompt as well as in total.

Tell us the challenge this post left you with.

A senior practitioner reads it and replies within one business day with what we would check first. No deck, no templates.