AnswerFix

How AnswerFix measures

What we ask, which engines answer, how we score your site, and what we found when we ran the same check five times in a row.

Which engines and models

AnswerFix queries three AI answer engines directly, using their own APIs rather than screen-scraping a chat window. ChatGPT answers come from GPT-6.1 Sol, which OpenAI describes as near-Astra performance for complex work at a lower cost, with web search turned on, so the model can read current pages rather than rely only on training data. Perplexity answers come from Perplexity Sonar, which is built specifically to search the web and cite sources for every claim. Paid plans add Gemini 4 Argon, which Google calls its flagship reasoning model, with Google Search grounding enabled, giving a third independent read on the same questions. We use the same settings for every business so results are comparable across the report. Earlier reports were answered by GPT-5 mini and Gemini 2.5 Flash, and every report names the model behind each answer.

How many questions, and how they are built

The free score asks ten local-intent questions, the kind a real customer would type: a mix of category plus city phrasing ("best plumber in Austin"), problem phrasing ("who fixes a burst water heater near me"), and comparison phrasing ("top rated dentist near downtown"). Paid plans expand the same pattern, to twenty five questions per location on Starter and fifty on Pro and Agency, adding more problem and comparison variants so the sample is large enough to show a stable trend rather than a single lucky or unlucky answer.

How we match your business, and what named means

We match a citation with conservative, whole-name matching against the exact business name you gave us in the intake form, not a fuzzy or partial match. Named means the engine's answer text contains that name, not merely a link, a category mention, or a competitor with a similar name. This deliberately under-counts rather than over-counts: we would rather report a false negative than tell you an AI engine recommended you when it recommended someone else with a similar name.

The 13 website checks

Alongside the AI answers, AnswerFix runs 13 checks directly against your live website: things like LocalBusiness schema, a canonical tag, an FAQ block engines can cite, clear entity statements about what your business does, and whether AI crawlers are allowed in robots.txt. Every one of these checks is verifiable instantly. View source or a schema validator will show you the exact same pass or fail we report, because we are reading your live HTML, not a cached or estimated copy.

What we measured about run-to-run variance

AI answers are not deterministic, so we tested our own consistency before asking you to trust it. We ran the identical check, on one unchanged business, five times in a row. Across those five runs the business was named in between 9 and 12 of 20 answers, the score moved with a standard deviation of about 3 points, and the top-ranked competitor named by the engines changed between runs. Nothing about the business or its website changed between those five runs, only the engines' own sampling. In practice this means a single score is a snapshot, not a verdict: look for a consistent direction across repeated checks, not a single number.

What we never do

We never invent facts about your business to fill out a report or a fix. We never publish or index a business's report; every report page is set to noindex and is only reachable by the link we give you. We never sell your data to anyone.

Limitations

Our queries carry no account history, no browsing history and no location beyond the city you name, so every business in a city is asked on the same footing and the numbers stay comparable. A signed-in customer standing on your street sees a personalised answer this check does not capture. Google AI Overviews are not measured; the FAQ below says why. Plan question counts and refresh schedules are on pricing.

Methodology questions

Why does my score change between runs?
Every AI engine samples its own answers, so the same question can come back differently minutes apart. In our own repeated test of one unchanged business, the score moved by a few points and the named competitor changed between runs. Read the score as a direction, not a fixed rank, and use the trend across several checks rather than any single run.
What does named mean?
Named means the engine's answer text includes your business by name, not just a link or a category mention. We check the raw answer text for a conservative, whole-name match against the business name you gave us, so a partial or unrelated match never counts as a citation.
Do you check Google AI Overviews?
Not yet. Overviews publish no stable, queryable interface, so nobody can sample them repeatably and quote the result back to you. We measure ChatGPT, Perplexity and Gemini, where the same question can be asked again and again and every answer shown to you in full. Overviews go in the moment a reliable method exists.

The method, applied at scale

We ran this same measurement across a sample of real businesses and published everything, including the dataset. See the AI visibility study for what the engines actually answered, which sources they cited, and where the numbers came from.

Run this method on your own business

The same engines, the same ten questions, the same 13 website checks. You get every answer word for word, the sources each engine used, and your top fix with the code to paste.

Free, no card, results in a few minutes.