Same Prompt. Same Day. 41% Overlap.

Greg McConnell

6 days ago

4 min read

Marketers are reporting their rank in AI answers. Rank needs a list that holds still, and the lists change between two runs of the same prompt.

Muck Rack's October 28 study of 207 million AI citations found that repeating a prompt on the same day returned the same source domains only 41% of the time on one major assistant.

The Lead

Marketing is leaving the ranked era. For twenty years a position meant something: page one, slot three, share of voice. AI answers publish no list that holds still. Muck Rack's analysis of 40 million AI responses measured same-day domain overlap at 41% on one assistant, 45% on Gemini and 94% on Perplexity. Only 212 domains were cited by all four providers tested. Meanwhile Semrush data shows AI Overviews on 88% of 400 tracked US commercial keywords in August, up from 45.75% a year earlier. The surface is growing while its output stays unstable. Gilroy's review of 174,745 UK B2B opinions found 80% of revenue leaders say AI search attribution is not validated. The consensus says to track your rank, as in SEO. The data supports tracking your odds. This week, ask whoever reports your AI visibility how many times each prompt was run.

The Take

The consensus: AI visibility is the new SEO, so find your rank and climb it. The evidence says otherwise. A rank belongs to a stable list, and these lists redraw themselves. Muck Rack's repeat runs overlapped 41% of domains on one assistant and 94% on another, so volatility varies by engine and a single check proves nothing. Paid placement cannot patch the gap: paid and advertorial content holds 0.3% of citations while earned and third-party sources supply more than 84%. Owned media, the asset most teams optimize, sits at 15.2%. A dashboard that reports one position from one run reports one draw from a distribution. Stop reporting positions. Report odds, with the run count behind them, and fund the earned coverage that moves them.

Rank needs a stable list. AI answers redraw the list on every run, so visibility is a probability.

Also Worth Knowing

Google tests ads in AI Mode with Personal Intelligence (Oct 16). Users who opt in to connect their apps may see ads personalized with those insights. Google says ads will not draw directly on Gmail or Photos and advertisers cannot access connected apps, and it has announced no new advertiser controls. Action: ask your Google rep in writing what reporting and exclusions exist before any budget enters the test.
Blocking one crawler can erase you from one engine. Muck Rack found 81% of major publishers blocking Claude's search crawler earned almost no Claude citations, against 12% of those allowing it. Visibility is a per-engine permission, and a one-line robots.txt change can revoke it without anyone noticing. Action: have your web team audit robots.txt this week for every AI search crawler and write down who approved each block.
Etsy tightens AI rules on listings (October). The first listing photo must be an unedited image of the real item, AI-generated functional guides are banned, and AI art stays allowed with disclosure. Marketplaces are writing provenance rules ahead of regulators, and brand content that ignores them risks delisting. Action: audit your marketplace listings for generated hero images and confirm each one meets the host platform's current policy.

Under the Radar

DataTrue's Consent Gap benchmark (Oct 28) tested 242 of the most-visited US sites in October. Of the 173 running ad tracking, 139 offered no obvious opt-out and 76% did not fully honor Global Privacy Control. On 20 of 24 sites with a clear reject button, ad companies were still contacted after the click. Trade press covered the privacy angle. The marketing angle is audience quality: every AI targeting model trained on this data inherits consents nobody verified. A model fed unverified consent produces unverifiable reach. Action: ask your CDP and DSP vendors for their Global Privacy Control honor rate in writing before renewal.

Frameworks & Vocabulary

Citation Odds: the share of repeated runs of a buying prompt in which your brand appears, reported per engine. It is the repeatable successor to Share of Prompt. Monday's meeting line: "Our Citation Odds on the pricing prompt are 35% on Gemini and 80% on Perplexity, across 20 runs each."

The Practitioner Move

Pick 10 buying prompts. Run each 20 times on each of 3 engines this week (600 runs). Log two fields per run: brand appears (yes or no) and the domains cited. Report Citation Odds per prompt per engine, with the run count. Re-run in 30 days. Treat any swing under 10 points as noise until a third run confirms it. Assign one owner and one date.

Think About This

A position you cannot reproduce is a sample of one.

Forward this to the head of SEO who still reports your brand's "position" in AI answers to the CMO.

About mktg.ai

mktg.ai is the Creative Intelligence System for modern marketing. The platform unifies every creative asset, channel, and KPI in one place, connecting creative performance to spend in real time so teams can act at the layer consumers actually experience. Features include Ask mktg.ai for natural-language queries against your marketing data, and Daily Alerts AI for automatic anomaly detection that surfaces performance issues without waiting for weekly reports. On a $5M media budget, mktg.ai customers typically recover over $100,000 by reallocating 15 to 20 percent of spend toward higher-ROI creative within months. Learn more at mktg.ai.