What does position mean in an AI answer?
AI answer position is the order in which a brand appears in the answer, not a universal search ranking. Record the first brand mention, the first cited page, and whether the brand appears at all. These are separate observations and should not be reduced to one score too early.
For example, a response might mention your company first but cite a competitor first. Another might cite your guide near the top while mentioning your brand later. A third might list your company without linking to a source. Each outcome has a different implication for what to change.
Use an ordinal position for each observable item. Position one means the first brand or source shown in the answer, position two means the second, and so on. Use “not present” when the item does not appear. Do not call position one a ranking, because answer layouts vary and some responses contain no ordered list.
Cituna measures whether seven engines mention and cite a brand: ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews and Google AI Mode. The useful starting record is therefore engine, prompt, brand position, cited-page position, and date.
For more context, read Branded vs Nonbranded AI Visibility: What to Measure.
Which buyer prompts should I measure first?
Start with the buyer questions where an answer could influence a shortlist, recommendation, or purchase decision. A small, stable prompt set is more useful than a large collection of loosely related questions because position only means something when the question stays comparable.
Choose prompts from sales calls, site search, customer support, product comparisons, and pages that already attract relevant organic visits. Include the wording buyers actually use, including prompts that name a problem, a category, a use case, or a comparison. Keep branded and nonbranded prompts in separate groups so a strong result for your own name does not conceal weak category visibility.
Write each prompt exactly as it will be tested. Record location, language, audience assumptions, and any requested constraints. “Best accounting software for a five-person agency in the UK” can produce a different answer from “best accounting software.” Those are different measurement units, not duplicate prompts.
Prioritise prompts with clear commercial consequences and enough specificity to diagnose an answer. The first question to check is not how many prompts you have. It is whether each prompt represents a decision your company wants to influence.
For more context, read How Ai Engines Decide Which Brands To Mention.
How do I run the same prompt consistently?
Run every comparison with the same prompt, engine, market, language, and collection conditions, then save the complete response and cited links. Consistency is necessary because AI answers can change with context, model updates, location, account state, and time.
Use a fixed prompt list and a defined collection schedule. Avoid adding follow-up questions when measuring the initial answer, because a conversation can change which brands and sources appear. Record whether the test was signed in, whether personalisation was active, and whether the engine displayed web citations or generated a response without them.
Keep the answer text rather than recording only a position number. The full response lets you verify whether the parser treated a heading, comparison table, citation block, or repeated mention correctly. It also preserves evidence when an engine changes its interface.
Rules and answer formats change, so document the engine and collection date with every observation. ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews and Google AI Mode do not necessarily expose sources or structure answers in the same way. Measurement is defensible when a later reviewer can reproduce the conditions and see the original evidence.
What exactly should I count as a brand position?
Count the first meaningful brand mention in the answer body, and record source position separately from brand position. A brand name in a citation, a heading, a comparison table, or the explanatory prose can have different significance, so label the location instead of treating every occurrence as equivalent.
Set a rule before reviewing results. For instance, count an explicit company name as a mention, but do not count a generic product category or an unnamed description. If the same brand appears several times, record its first position and note repeated appearances separately when they add context. Ignore navigation labels and unrelated footer text unless the engine presents them as part of its answer.
For citations, count the first source that clearly points to a page from the brand’s domain. A domain appearing in a source list is evidence of citation, even if the prose never names the company. A brand mentioned in prose without a link is a mention without a citation.
The key check is whether the position rule produces the same result when two people review the same answer. If not, clarify the rule before comparing engines or reporting movement.
How should I record an answer where my brand is absent?
Record an absent brand as “not present,” never as a low position such as the last place. Absence is a different outcome from appearing below several competitors, because the corrective action may involve relevance, evidence, or source selection rather than improving prominence.
Use a result record with at least four states: mentioned and cited, mentioned but not cited, cited but not mentioned, and neither mentioned nor cited. Add the brand position and source position when applicable. This prevents a single visibility percentage from hiding whether an engine knows the company, trusts a page, or simply omitted it in that response.
Preserve the complete answer for absent cases. An omission may be caused by a narrow prompt, a source set that excludes your site, a competitor’s stronger explanatory page, or an answer that does not attempt to name providers. The answer text gives the reviewer something to diagnose instead of inviting speculation.
Treat absence as a result for that exact prompt and engine, not proof that the brand is invisible everywhere. Repeating the same test under changed conditions can reveal volatility, while the fixed record still provides a reliable baseline.
Should I combine positions across engines?
Combine engine results only after preserving the engine-level positions, and use a distribution rather than a single blended rank. A brand in position one on one engine and absent on another has a different risk profile from a brand consistently in position three across both.
Create a row for every prompt and engine. Then compare the share of tests with a first mention, any mention, a first-party citation, and no appearance. Median position can describe tests where a brand appears, but it must not treat absent results as ordinary low ranks. Keep the denominator visible so a promising median does not conceal frequent omission.
Compare like with like. Google AI Overviews and Google AI Mode may answer within a search results page, while ChatGPT, Perplexity, Gemini, Claude and Grok may use different response and source formats. A cross-engine summary is useful for direction, but the individual engine view tells you where the problem occurs.
Use a decision rule: prioritise a prompt when it has commercial importance, repeated absence or weak citation, and a clear page or competitor pattern. Do not prioritise a small position change that has no stable evidence or no plausible action attached to it.
When is manual checking enough, and when is tracking better?
Manual checking is enough for validating a small prompt set, while recurring measurement requires a consistent tracking method. Manual review is valuable at the start because it exposes answer structures and edge cases that a spreadsheet or parser may misclassify.
For a manual check, copy the prompt, record the engine and conditions, save the full answer, mark brand and citation positions, and write one diagnostic note. Repeat the same review later rather than comparing an old screenshot with a new answer that used different conditions. A spreadsheet can work when the set is small and the reviewer applies stable rules.
A tracking system becomes more useful when daily or repeated collection would otherwise create missed observations, inconsistent transcription, or no clear view of competitor sources. Cituna tracks mentions and citations across ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews and Google AI Mode every day. It also shows which competitors and pages those engines cite, joins the answers to Google Search Console data, and provides SEO, AEO and GEO fixes.
The choice is not manual versus automated forever. Use manual reviews to audit the measurement rules, then use repeatable tracking to spot patterns and return to the original answers before changing pages.
What should I change when position is weak?
Change the element that explains the observed failure, not the page with the lowest position by default. A brand that is mentioned but not cited has a different problem from a brand that is cited by a relevant page but appears after competitors.
If the answer omits the brand and cites unrelated sources, first check whether the prompt matches the page’s subject and whether the site clearly answers the buyer’s question. If competitors are mentioned first and their pages are cited, compare the evidence those pages provide, including definitions, use-case detail, limitations, and current information. If your brand is cited but the answer describes it inaccurately, fix the source page’s clarity before chasing more mentions.
Use Google Search Console data to connect an AI answer observation with organic query and page evidence, but do not assume organic clicks automatically determine answer position. AI engines may select sources for reasons that differ by query and engine.
After making one meaningful change, rerun the same prompt set and compare the affected engine, position type, and citation state. The practical sequence is diagnose, change one relevant source or message, retest, and keep the original answer for comparison. That sequence prevents position measurement from becoming a vague visibility score.
Related reading
Sources consulted
- OpenAI platform documentation (platform.openai.com)
- Perplexity API documentation (docs.perplexity.ai)
- Google Search documentation (developers.google.com)
- Google Search Central (support.google.com)
Drafted with AI assistance from our own research and Search Console data, and reviewed by Rahul A before publishing. Rules and prices change; check the linked official source before you act.