ChatGPT Core Updates: Chatbots Have Algorithm Changes Too

Everyone in SEO knows Google Core Updates: overnight, the visibility of domains shifts. Our data now shows the same pattern in ChatGPT. On May 23, 2026, the switch from GPT-5 mini to GPT-5.5 shifted citations in German-language ChatGPT responses by 47% within 48 hours.

Note: The data in this analysis is based on German-language queries and the German market. Similar patterns have, however, been observed in other markets as well.

What Google updates and ChatGPT updates have in common

A Google Core Update doesn’t change search itself, but how Google evaluates sources. The typical marker: strong shifts in rankings, many domains moving in both directions simultaneously, triggered by a datable event.

ChatGPT doesn’t have list rankings, but it has an equivalent: citations. Per response, ChatGPT cites an average of 25–35 sources. Which domains those are reflects what the model judges as relevant and authoritative. If that evaluation shifts systematically, that’s the core update equivalent.

With the SISTRIX AI Visibility Index, this shift can be measured to the day and across a large sample. The methodology is documented separately. For the May 2026 study, we evaluated 38 daily samples of 100,000 German-language ChatGPT responses each — a total of 3.8 million responses and over 100 million source mentions.

Two observed update events in May 2026

During the observation period, two separate shift events are visible in the SISTRIX data. One of them is methodologically attributable and needs to be distinguished from the actual ChatGPT update:

Event 1, May 7–11, 2026 (ChatGPT image embeddings): OpenAI rolls out a layout change where ChatGPT responses contain significantly more inline embedded images via images.openai.com. The share of responses with at least one inline element rises from ~15% to ~35%. Citations in the classic sense barely change.

Event 2, May 22/23, 2026 (GPT-5.5 rollout): Within 48 hours, the domain distribution in German-language responses shifts by 47% (comparison value in the preceding days: 1–2%). At the same time, the average number of sources per response drops from 30.9 to 28.4. The switch was verified via daily sample queries to ChatGPT using the prompt “which model are you” — up to May 22, the system responded “GPT-5 mini”; from May 23 onward, “GPT-5.5”.

Event 2 is what we refer to as the ChatGPT Core Update.

Is this really an “algorithm update”? Not quite mechanically: a Google Core Update recalibrates the ranking system with the same index. With ChatGPT, citations are determined by the interplay of the retrieval backend (which sources are fetched) and the model itself (what it judges as relevant and cites). A version jump like GPT-5 mini to GPT-5.5 completely replaces the second component — it’s more like a new ranking core than a reweighting. At the impact level that matters for GEO, however, it behaves identically to a core update: datable, systemic, with clear winners and losers.

Important for context: We are measuring a correlation. The model identifier switched on May 23 from “GPT-5 mini” to “GPT-5.5”, and simultaneously the citation distribution shifted. Whether the model change alone was the cause, or whether the retrieval or prompt layer was also changed in parallel, cannot be determined from the outside.

Winners and losers of the GPT-5.5 rollout

We compare the citation distribution in the four days before the update (May 18–21) with the four days after (May 26–29). Data basis: 800,000 ChatGPT responses.

The top winners (Germany)

ChatGPT Core Update - Winners, Germany
DomainAmount before (per 10k)Amount afterΔ absoluteΔ relative
reddit.com1185318860+7007+59 %
welt.de427850+424+99 %
mapbox.com449822+373+83 %
justwatch.com43314+271+624 %
bild.de244447+203+83 %
legalclarity.org9191+182+1928 %
dazn.com47226+180+383 %
techradar.com409562+153+37 %
tomsguide.com196334+138+70 %
rnd.de32162+130+405 %
chip.de534662+128+24 %
openstreetmap.org152279+127+83 %
sky.de76195+119+157 %
netzwelt.de142257+114+80 %
computerbild.de163275+112+68 %
fernsehserien.de41145+103+250 %
testbericht.de28131+103+362 %
faz.net83185+102+124 %
joyn.de59161+101+171 %
kicker.de25112+88+357 %

The top losers (Germany)

ChatGPT Core Update - Losers, Germany
DomainAmount beforeAmount afterΔ absoluteΔ relative
youtube.com80316548-1483-18 %
indeed.com27891473-1315-47 %
wikipedia.org70996091-1008-14 %
tripadvisor.com1503706-798-53 %
facebook.com35822976-606-17 %
glassdoor.com1391875-516-37 %
glassdoor.co.uk942455-487-52 %
kununu.com1055574-481-46 %
rome2rio.com650262-388-60 %
ziprecruiter.com922610-312-34 %
google.com1279993-286-22 %
classic.com29880-219-73 %
yahoo.com866704-162-19 %
linkedin.com689538-151-22 %
expedia.com24898-150-60 %
tripadvisor.co.uk286139-147-51 %
mobile.de22074-146-66 %

The industry pattern: localization vs. international aggregators

The individual domain movements add up to a clear pattern in the aggregate.

Three clusters gained:

Established German mainstream media: welt.de +99%, faz.net +124%, bild.de +83%, chip.de +24%, computerbild.de +68%, netzwelt.de +80%, rnd.de +405%. Also German TV/streaming: sky.de +157%, joyn.de +171%, dazn.com +383%, fernsehserien.de +250%, kicker.de +357%.

Reddit: With +59% and +7,007 citations per 10,000 responses, Reddit is the biggest winner, both in absolute and relative terms. Notably, Reddit was already the market leader in the citation landscape.

Specialized tools and knowledge sources: mapbox.com +83%, openstreetmap.org +83%, justwatch.com +624%, testbericht.de +362%, legalclarity.org as a new entrant.

Losers:

International aggregator and comparison platforms: Indeed −47%, Glassdoor (all country versions) −37% to −52%, Tripadvisor (.com & .co.uk) −51% to −53%, Expedia −60%, Rome2rio −60%, ZipRecruiter −34%, Mobile.de −66%, Classic.com −73%.

Global tech platforms: YouTube −18%, Facebook −17%, Google.com −22%, LinkedIn −22%, Wikipedia −14%, Yahoo −19%.

German comparison/career portals: Kununu −46%, Glassdoor.de −50%.

In aggregate, a clear localization emerges: for German-language queries, GPT-5.5 cites originally German sources more frequently – both established premium publishers and German service brands. International aggregators that previously dominated many query types are losing ground, as are generalists like Wikipedia, YouTube, and Facebook. Whether this shift is a deliberate goal of the model cannot be derived from the data.

The only notable exception is Reddit, which continues to gain share against the general localization trend. A plausible explanation: Reddit content is predominantly in English, but user-generated discussions covering all topics continue to be rated as highly relevant across language lines.

How significant was the shift?

Citation drift comparison
DateCitation DriftClassification
Typical weekday (May 2026)1–2 %Statistical noise
May 7–11 2026 (Inline layout)2–2.5 %Subtle update
May 22/23 2026 (GPT-5.5)47 %Core update level

A citation drift of 47% means: almost half of all citations go to different domains after the update than before. For comparison: a typical Google Core Update shifts the share of top domains at a similar magnitude when measured by the visibility index; usually 5–10% of the top 100 domains land in new top rankings on a day-over-day basis.

The GPT-5.5 rollout is therefore not a gradual transition, but a step event on the scale of a significant Google Core Update.

What does this mean for GEO and SEO?

Three implications from the data:

  1. For German publishers and service brands: a clear visibility gain. Those working with established German-language content and brand authority in the German market have been cited more frequently as a source in ChatGPT responses since May 23. Welt, FAZ, Bild, Chip, Computerbild — all have moved closer to the shares these brands already hold in classical Google SEO.
  2. For international aggregators: a visibility drop. Indeed, Glassdoor, Tripadvisor, Expedia, Rome2rio — platforms that gained market share through English-language content and global scale in the early weeks of ChatGPT are now giving back significantly. For this category, the next step is localized content and brand authority in the German-speaking market.
  3. GEO strategy needs the same monitoring as SEO. When ChatGPT and other generative search systems roll out updates of this magnitude, brands need to monitor them just as they would Google Core Updates. The movements are real, datable, and relevant for organic visibility. SISTRIX Prompt Tracking provides the data basis for this.

Conclusion

May 23, 2026 is the first clearly documented ChatGPT Core Update. In 48 hours, the switch to GPT-5.5 redistributed almost half of all citations, with a clear pattern: German publishers and service brands up, international aggregators down.

For GEO, this means: generative search systems have algorithm updates with winners and losers, and anyone who wants to be visible there needs the same reflexes as in classical SEO. ChatGPT is no longer a technical novelty – it’s a search system like any other.

Related posts